Pattern and process in the evolution of the sole dioecious member of Brassicaceae

Background Lepidium sisymbrioides, a polyploid New Zealand endemic, is the sole dioecious species in Brassicaceae and therefore the closest dioecious relative of the model plant Arabidopsis thaliana. The attractiveness of developing this system for future studies on the genetics of sex determination prompted us to investigate historical and developmental factors surrounding the evolution of its unisexual flowers. Our goal was to determine the evolutionary pattern of polyploidization of L. sisymbrioides and the timing and process of flower reproductive organ abortion. To that end, we used a combination of phylogenetics to place this species within the complex history of polyploidization events in Lepidium and histology to compare its floral ontogeny to that of its closest hermaphroditic relatives and to A. thaliana. Results Using a nuclear locus (PISTILLATA), we reconstructed the gene tree among Lepidium taxa and applied a phylogenetic network analysis to identify ancestral genomes that contributed to the evolution of L. sisymbrioides. Combining this phylogenetic framework with cytological and genome size data, we estimated L. sisymbrioides as an allo-octoploid resulting from three hybridization events. Our investigations of flower development showed that unisexual flowers appear to abort reproductive organs by programmed cell death in female flowers and by developmental arrest in male flowers. This selective abortion occurs at the same floral developmental stage in both males and females, corresponding to Arabidopsis stage nine. Conclusions Dioecy in Brassicaceae evolved once in L. sisymbrioides following several allopolyploidization events, by a process of selective abortion of reproductive organs at intermediate stages of flower development. Different developmental processes, but similar timing of abortions, affect male versus female flower development. An increased understanding of how and when reproductive organs abort in this species, combined with our estimates of ancestral genome contributions, ploidy and genome size, lay the foundation for future efforts to examine the genetic mechanisms involved in the evolution of unisexual flowers in the closest dioecious relative of the best studied model plant. Electronic supplementary material The online version of this article (doi:10.1186/2041-9139-5-42) contains supplementary material, which is available to authorized users.


Background
The family Brassicaceae contains approximately 338 genera and 3,709 species [1,2] and includes the model plant Arabidopsis thaliana, which has a floral morphology representative of the vast majority of the family. Only 5% of genera within Brassicaceae show deviations from the basic floral plan of four sepals, four petals, six stamens (four medial and two lateral), and two fused carpels [3]. One of these genera that diverge from the basic Brassicaceae floral morphology, Lepidium (230 species [1]), is widely distributed in temperate and subtropical areas [4]. Earlydiverging lineages in the genus comprise outcrossing diploid species from the Old World that exhibit the basic floral plan of the family, whereas derived lineages tend to be selfing allopolyploids from the New World, Australia, and New Zealand with reduced flowers (that is, fewer stamens and/or reduced petals) [5]. Among the latter, the New Zealand endemic Lepidium sisymbrioides is the only dioecious species in the whole Brassicaceae [4,[6][7][8][9][10]. Staminate flowers of L. sisymbrioides consist of four to six stamens and a reduced ovary [10,11], whereas carpellate flowers have three to seven staminodes and a functional pistil with style and stigma [10]. Nonfunctional reproductive organs of unisexual flowers of L. sisymbrioides have been loosely described as 'abortive' [8,11], but the exact timing and process of the abortions remain unknown.
Incongruence between phylogenetic trees using nuclear versus chloroplast DNA regions suggests reticulate evolution within the genus [12]. It appears that hybridization, followed by whole genome duplication, resulted in predominantly allopolyploid Lepidium species in the Americas, Australia, and New Zealand [5]. These hybridization events may have contributed to the reduced stamens and petals observed in these species [5], in which case dioecy could represent another example of organ reduction.
In angiosperms, dioecy often follows polyploidization, presumably due to chromosomal rearrangements that facilitate the evolution of sex chromosomes or the breakdown of gametophytic self-incompatibility, followed by inbreeding depression [13][14][15]. Correlations between island habitat and dioecy are also common, through selection for outcrossing in small, colonizing hermaphroditic populations, [16]. In fact, New Zealand taxa in general show a higher incidence of gender dimorphism compared to their continental sister taxa [17,18]. Combined evidence for reticulate evolution and polyploidy in New Zealand Lepidium [5,12] suggests that L. sisymbrioides may also be an allopolyploid. We therefore hypothesize that this species represents another case of dioecy evolving in an island species, following hybridization and polyploidization.
The lineages containing Lepidium and Arabidopsis diverged from each other relatively recently, approximately 35 million years ago (mya) [19]. Lepidium sisymbrioides, therefore, offers the potential to uncover the genetic mechanisms involved in the evolution of unisexual flowers by being the closest dioecious relative to the most thoroughly investigated model plant. Determining the developmental stage and process of reproductive organ abortion should facilitate the identification of candidate genes involved in the evolution of unisexual flowers in this species as genes involved in sporogenesis and gametogenesis have been identified in Arabidopsis [20,21]. Six developmental processes leading to reproductive organ abortion in unisexual flowers are recognized: cell death, programmed cell death, parenchymatization, arrest of development, change in timing of otherwise normal developmental events, and inviable pollen [22]. Identifying which of these processes contributes to the development of male and female flowers in L. sisymbrioides, as well as estimating this species' genome size and ploidy history, will facilitate future efforts to uncover the genetic mechanisms involved in the evolution of dioecy.
The overall goal of this study was to investigate the pattern of polyploidization and the developmental processes underlying the evolution of separate sexes in L. sisymbrioides, the sole dioecious member of Brassicaceae. To that end, we 1) identified ancestral genomes within L. sisymbrioides and close relatives, 2) estimated ploidy and genome size for L. sisymbrioides and close relatives and 3) investigated the timing and process of organ abortion by comparing its floral development to that of its close hermaphroditic relatives and to Arabidopsis thaliana.

Sampling methods
Three subspecies of L. sisymbrioides were originally recognized: kawarau (Petrie) Thell., matau (Petrie) Thell., and sisymbrioides. All have been listed as nationally endangered because of a steep reduction in their distribution and abundance [23] and are difficult to sample. We sampled L. sisymbrioides subsp. sisymbrioides only because, despite reported habitat and morphological differences, all three subspecies are closely related [10].
In addition to published DNA sequences of Lepidium available from GenBank (Appendix 1), we sampled L. sisymbrioides subsp. sisymbrioides and three hermaphroditic close relatives endemic to New Zealand, L. kirkii, L. naufragorum and L. tenuicaule [9,10,12,24,25]. We sampled L. kirkii from an herbarium specimen (Appendix 1) and the remaining three species from cultivated accessions at the University of Washington (UW) greenhouse from wild-collected seed provided by P. Heenan (Landcare Research, Lincoln, New Zealand). Voucher specimens are listed in Appendix 1.

Molecular methods
Because we were primarily interested in the polyploidization history of L. sisymbrioides and its close relatives (L. kirkii, L. naufragorum, and L. tenuicaule), we investigated reticulation events using the single-copy nuclear gene PISTILLATA (PI). PI had been previously used to detect reticulation among other Lepidium taxa [5] and therefore sequences were readily available (Appendix 1). Genomic DNA was extracted from one to two accessions for each of our four study species using the FastDNA Kit (MP Biomedicals, Solon, OH, USA) for cultivated accessions or following the protocol of Hughey et al. [26] for the herbarium specimen. We amplified and sequenced the first intron of PI using PI-ITF and PI-ITR primers [5]. Polymerase chain reaction conditions were 95°C for 2 min, followed by 35 cycles of 94°C for 30 s, 60°C for 1 min, and 72°C for 1 min, with a final extension step at 72°C for 5 min. Amplified DNA was purified using ExoSAP-IT (USB Corporation, Cleveland, OH, USA).
To distinguish among allelic variants, PCR products were cloned into the pCRII or pCR2.1 vector using the TA Cloning Kit (Invitrogen Corporation, Carlsbad, CA, USA). Plasmids were extracted using the FastPlasmid Mini Kit (5 Prime Inc., Gaithersburg, MD, USA). Three to 27 positive clones were sequenced per accession (UW Biochemistry DNA Sequencing Facility or GENEWIZ, Seattle, WA, USA) for a total of 20 to 35 clones per taxon.

Phylogenetic analyses
PISTILLATA sequences were edited in Sequencher 4.9 (Gene Codes Corporation, Ann Arbor, MI, USA). We incorporated sequences we generated from our four study species plus available sequences from other taxa in Gen-Bank to the entire alignment of the PI first intron, provided by J. L. Bowman [5]. We then aligned all sequences manually using MacClade 4.08 [27]. Ambiguously aligned regions were excluded from subsequent analyses. Our data set is available through TreeBASE (http://purl.org/phylo/ treebase/phylows/study/TB2:S11886).
In order to detect whether PCR-mediated recombination had occurred among the PI copies within a species, we checked for recombination using RDP4 [28]. The PI alignment was analyzed by the automated exploratory recombination analysis, which employs eight different recombination methods: RDP [29], BootScan [30], GEN-ECONV [31], MaxChi [32], Chimaera [33], SiScan [34], 3Seq [35], and LARD [36]. Analyses were run under the default general settings but as linear sequences and disentangling overlapping signals. The default settings for each method were used except for the following models: Felsenstein 1984 for BootScan and reversible process for LARD. Five recombinant sequences were identified and removed from the PI alignment before subsequent phylogenetic analyses.
We reconstructed the phylogeny for the Lepidium PI data set using Bayesian and likelihood analyses. For taxa that had multiple clonal PI sequences, we chose one clonal sequence from each monophyletic group of sequences representing a given taxon that was recovered in a 50% majority rule consensus tree from preliminary Bayesian analyses of all clones and that was representative of the majority of clones from a group. All other clonal sequences not forming monophyletic groups with other clones from the same taxon were included in the final analyses (Appendix 1).
For Bayesian and likelihood analyses, the model of evolution for the PI data set was determined by jModelTest 2.1 [37,38]. The model selected under the Akaike Information Criterion [39] was TVM + I + Γ. We specified L. phlebopetalum and L. perfoliatum as outgroups, as previously identified in various studies [5,9,25,40].
Bayesian analyses were conducted in MrBayes 3.2.2 [41,42] via the CIPRES Science Gateway 3.3 [43]. We used default priors of no prior knowledge for the parameters of the model. Bayesian analyses were conducted with three independent Markov Chain Monte Carlo [44] analyses of 10 million generations each. Metropolis coupling for each analysis was conducted under the default settings. Convergence was determined when the average standard deviation of split frequencies remained less than 0.01. The first 10% of trees was discarded before convergence. The remaining trees from each run were pooled to construct a 50% majority rule consensus tree to obtain posterior probabilities (pp) and visualized with FigTree 1.4 [45].
Likelihood analyses were conducted in GARLI 2.0 [46]. Analyses were run under the default settings and included five search replicates to determine the maximum likelihood tree. To assess the reliability of clades in the resulting likelihood tree, we conducted 1,000 nonparametric bootstrap (bs) replicates [47] in GARLI. Bootstrap replicates were conducted under the above settings, but included one search replicate and 10,000 generations as the first part of the termination condition. Bootstrap trees were summarized with SumTrees 3.3.1 [48] and visualized with FigTree.
Since multiple PI copies in polyploid Lepidium taxa had been previously ascribed to allopolypoidy [5], we wanted to identify potential hybridization events leading to the evolution of dioecy in L. sisymbrioides. To facilitate visualization of potential ancestral genomes that contributed to the evolution of L. sisymbrioides and its close relatives (L. kirkii, L. naufragorum, L. tenuicaule), we conducted network analyses using the 50% majority rule consensus tree from Bayesian analyses of the PI data set as a multilabeled tree (MUL tree). The MUL tree was imported into Dendroscope 3.2.10 [49] and transformed into a phylogenetic network using the Huber et al. [50] algorithm, which minimizes the number of hybridization nodes.

Cytology
Chromosome counts were obtained from pollen mother cells (PMCs) from freshly collected floral buds from one to two cultivated accessions of L. sisymbrioides and L. tenuicaule using a modified protocol of Kato [51]. Floral buds were treated according to the protocol of Matsushita et al. [52] and Wright et al. [53], modified with an N 2 O treatment for 3 hours at 206 PSI and an enzyme digestion for 3 hours. PMCs were mounted in VECTASHIELD with DAPI (Vector Laboratories, Burlingame, CA, USA), observed and photographed using a Nikon Microphot-FX microscope (Nikon Instruments, Inc., Melville, NY, USA) and a Retiga 1300 monochrome camera (QImaging, Surrey, British Columbia, Canada).

Genome size estimation
Three accessions from three species cultivated at the UW greenhouse (L. naufragorum, L. sisymbrioides, and L. tenuicaule) were analyzed to obtain relative holoploid genome size, expressed as a 1C-value as defined in [54]. Nuclei were extracted from fresh leaf tissue and combined with chicken erythrocyte nuclei (CEN singlets, BioSure, Grass Valley, CA, USA) before staining with propidium iodide and analyzed with a flow cytometer, as outlined in Davison et al. [55]. CEN, with a 1C-value of 1223 Mbp or 1.25 pg [56], were used as an internal calibration standard. Animal standards have been discouraged by some authors for plant studies because (1) they cannot account for the huge range of plant genome sizes, (2) their nuclei structure may be different from plant nuclei, and (3) their precise genome size is unknown [57]. However, for our purposes, the genome size of CEN falls within the range of Lepidium genome size estimates. Additionally, our goal was to produce relative holoploid genome size estimates, since absolute estimates are not feasible due to the lack of complete genome coverage in most model taxa because of repetitive regions in the genome [55,57].
Samples were analyzed on a FACScan flow cytometer (Becton, Dickinson and Company, Franklin Lakes, NJ, USA) with FlowJo software (Tree Star, Ashland, OR, USA) at the UW Department of Immunology Cell Analysis Facility. The 2C median nuclear peak of propidium iodide fluorescence in Lepidium samples was compared to that of the CEN standard to estimate the 2C nuclear DNA content of Lepidium in Mbp, then converted to pg using the equation from Dolezel et al. [58].
The relative timing of floral organ initiation and growth in Lepidium taxa had been previously shown to be comparable to that of its model Brassicaceae relatives (that is, Arabidopsis thaliana and Brassica napus), with the exception of petal growth and stamen number [59,60]. Therefore, floral developmental stages for our three Lepidium study species were designated by the 13 characterized stages of the closely related model Arabidopsis thaliana [20,61,62].
For histological observations, inflorescences were fixed in FAA, then dehydrated through an ethanol series ending in Citrisolv (Fisher Scientific, Kent, WA, USA), embedded in Paraplast Plus (McCormick Scientific, LLC, St. Louis, MO, USA), and sectioned (5 or 8 μm) according to the protocol of Kramer [63]. Slides were deparaffinized with CitriSolv, hydrated through an ethanol series, stained in 1% Safranin O for 24 hr [64] and counterstained with 0.5% Fast Green FCF for 30 sec or stained in 0.05% Toluidine Blue O in dH2O for 1 to 2 min, and dehydrated through an ethanol series ending in Citri-Solv. Histological sections were mounted in Cytoseal™ 60 (Richard-Allan Scientific, Kalamazoo, MI, USA) and observed using a Leica TCS SP5 II laser scanning confocal microscope (Leica Microsystems Inc., Buffalo Grove, IL, USA) with an excitation of 488 nm and an emission of 500 to 560 nm for Safranin O and an excitation of 561 nm and an emission of 625 to 690 nm for Fast Green FCF or using a Leitz Orthoplan 2 microscope (Ernst Leitz, Midland, Ontario, Canada) and photographed with a MicroPublisher 3.3 Real-Time Viewing camera (QImaging, Surrey, British Columbia, Canada).

TUNEL assays
We conducted TUNEL assays to determine whether programmed cell death (PCD) was occurring in aborted stamens from L. sisymbrioides female flowers. Paraffinembedded tissue sections were prepared as outlined above from female L. sisymbrioides and male L. sisymbrioides and hermaphroditic L. tenuicaule for comparison. We used the DeadEnd Fluorometric TUNEL System (Promega Corporation, Madison, WI, USA) according to manufacturer's instructions, including positive controls, and washed slides in PBS containing 0.1% Triton® X-100 and 5 mg/ml of BSA after terminating reactions to reduce background as recommended. Slides were mounted in VECTASHIELD with DAPI, except for negative control slides that were untreated and mounted in Cytoseal™ 60. Slides were observed using a Leica TCS SP5 II laser scanning confocal microscope using an excitation of 405 nm and an emission of 430 to 550 nm for DAPI and an excitation of 488 nm and an emission of 500 to 535 nm for fluorescein.

PISTILLATA gene duplication history suggests allopolyploidy in Lepidium sisymbrioides and relatives
In order to identify hybridization events leading to the evolution of the dioecious species L. sisymbrioides and its closest hermaphroditic relatives, L. kirkii, L. naufragorum and L. tenuicaule, we amplified and sequenced the first intron of the single-copy nuclear gene PI from these species and aligned them to other Lepidium sequences available in GenBank or unpublished (provided by J. L. Bowman; Appendix 1). Phylogenetic analyses recovered five, strongly supported clades (A1-D; pp ≥0.99, bs ≥94%; Figure 1) representing five major copies of the PI intron from American, Australian, and New Zealand (AANZ) taxa. Multiple copies of the PI intron within a taxon were previously suggested as representing multiple genomes from allopolyploid hybridization [5]. Clades A1 and A2 were denoted here because they had been previously recognized as a single clade ' A' , but with low support [5]. In our analysis, these two clades are strongly supported as distinct (pp = 1.00, bs = 100%) and indicative of two separate genomes, as evidenced by sequences from our four study species in both clades ( Figure 1, colored dots). Therefore, we found at least four distinct copies of the PI intron in L. kirkii  Figure 1). In addition, multiple sequences of L. kirkii, L. naufragorum, L. pseudohyssopifolium, and/or L. tenuicaule within clades A1, C, and D ( Figure 1) suggest that hybridization, gene duplication and/or allelic divergence are at play. None of the New Zealand taxa studied fell into the fifth clade B, which consists entirely of American taxa. Our results therefore suggest that all four New Zealand species are allopolyploids (and potentially allo-octoploids at minimum), originating from at least four divergent genomes (represented by clades A1, A2, C, and D).
We further used the Bayesian majority rule consensus tree from the PI data set (Figure 1) to estimate a phylogenetic network to aid in the identification of hybridization  Four PI clades previously identified from a number of taxa within the genus are denoted as A-D (after [5]). Clade 'A' is not monophyletic in our study, and new clades identified by our study are denoted as A1 to A2. Sequences from Lepidium taxa that were generated in our study are in bold and indicated by a colored dot: L. kirkii (yellow), L. naufragorum (purple), L. sisymbrioides (red), L. tenuicaule (green). Four PI copies from L. sisymbrioides are indicated by arrows. Posterior probabilities ≥0.90 and maximum likelihood bootstrap values >50% are displayed above and below branches, respectively.
nodes and potential ancestral genomes contributing to our study species (Figure 2). According to the network, our sampling includes 21 allopolyploid Lepidium taxa ( Figure 2, tree branches originating from curved lines), which are confirmed polyploids from the literature [5] and this study (L. sisymbrioides, L. tenuicaule). The remaining 31 taxa in our analyses do not show evidence of reticulation, and this may be due to diploidy, autopolyploidy, or gene loss. The evidence suggests that Lepidium sisymbrioides is derived from four distinct ancestral genomes (Figure 2, red lines): (1) a hybrid between (a) a descendant from the common ancestor of the L. monoplocoides group and (b) the common ancestor of the group that includes L. pseudotasmanicum and L. hyssopifolium (strong support), (2) a descendant from the common ancestor of the L. vesicarium group (low support), and (3) a descendant from the common ancestor of the L. dictyotum and L. quitense group (strong support). Biogeographically, the contribution of these genomes to L. sisymbrioides implies hybridization among Australian and New Zealand (ANZ) taxa (1a and 1b, above), followed by hybridization with American (3) and potentially (with low support) Asian (2) species. The other three New Zealand study species, L. kirkii, L. naufragorum, and L. tenuicaule, show contributions from four, five, and four distinct ancestral genomes, respectively ( Figure 2). Of these three close hermaphroditic relatives, L. sisymbrioides shares the most reticulation history with L. kirkii, followed by L. tenuicaule, then L. naufragorum ( Figure 2). Ancestral genomes contributing to our study species L. kirkii, L. naufragorum, L. sisymbrioides, and L. tenuicaule are indicated in colors matching Figure 1: yellow, purple, red, and green lines, respectively. Four distinct ancestral genomes (red) contributing to L. sisymbrioides are numbered on the right (1a, 1b, 2 and 3) with their respective biogeographical origin: Am, America; As, Asia; Au, Australia; ANZ, Australia/New Zealand.
Cytological observations reveal octoploidy in dioecious L. sisymbrioides and its hermaphroditic relative L. tenuicaule In order to confirm our results from the PI data set, we conducted chromosome counts in PMCs. Seed of L. kirkii was not available, so its chromosome number remains unknown.
Both L. sisymbrioides and L. tenuicaule had 2n = 64 chromosomes (Figure 3), corresponding to a ploidy of 8x (x = 8 is the base chromosome number for the genus [2,4]). Octoploidy in these two species is consistent with having four distinct copies of the PI intron as shown by our phylogenetic and network analyses (A1, A2, C, D, Figures 1, 2). These four copies would therefore represent four distinct diploid genomes in these species' history of hybridization and polyploidization events.

Genome size estimations confirm ploidy estimates in Lepidium sisymbrioides and relatives
We examined holoploid genome size by calculating 1C-values for three of our New Zealand Lepidium study species to confirm our estimates of ploidy and to inform future genomic sequencing plans. Lepidium sisymbrioides and L. tenuicaule had similar holoploid genome sizes of 0.63 and 0.66 pg, respectively, whereas L. naufragorum's holoploid genome size was slightly over double that of the other two species at 1.41 pg (Table 1). Material from which the holoploid genome size of L. naufragorum was obtained had published chromosome counts from the same population, indicating a ploidy of approximately 18x [65]. In conclusion, our holoploid genome size estimations are consistent with L. sisymbrioides and L. tenuicaule both being 8x and with L. naufragorum being 18x, more than double the ploidy of the former two species.
Female and male flowers of Lepidium sisymbrioides abort reproductive organs at comparable developmental stages but due to different processes In order to assess the developmental stage and process of abortion of reproductive organs in L. sisymbrioides, we examined floral morphology and ontogeny of this species in comparison to the two closest hermaphroditic relatives available, L. naufragorum and L. tenuicaule. Since floral developmental stages of hermaphroditic Lepidium species are comparable to the A. thaliana ontogenetic staging [20,61,62], we cross-referenced to this system for convenience and reproducibility. Flower morphology of L. naufragorum and L. tenuicaule differed from A. thaliana in petal size, number and arrangement of stamens, and ovule number. Lepidium naufragorum flowers had petals approximately as long as sepals and two lateral and two medial stamens ( Figure 4A), whereas L. tenuicaule flowers had highly reduced petals, unnoticeable to the naked eye, and four medial stamens ( Figure 4B). All Lepidium taxa produced a single ovule per locule.
In contrast to L. naufragorum and L. tenuicaule, both male and female flowers of L. sisymbrioides generally exhibited six stamens (two lateral and four medial; Figure 4C-F) with a few exceptions where only four medial ( Figure 5O) or five stamens were found [see Additional file 1]. Petals were reduced (that is, shorter than sepals; Figure 4C-F, arrowheads), and four to six nectaries were present among the stamen filaments in both sexes ( Figure 4D, F, asterisks; [see Additional file 1]). In staminate flowers, the gynoecium arrested its development at intermediate stages, after differentiation of the anther locules ( Figure 4C) and remained as a pistillode while stamens expanded normally ( Figure 4D), as in L. naufragorum and L. tenuicaule ( Figure 4A-B). In young carpellate flowers, stamens and carpels looked normal ( Figure 4E). In later stages, however, stamen development was visibly arrested resulting in staminodia, whereas the A B gynoecium developed normally ( Figure 4F) as in L. naufragorum and L. tenuicaule ( Figure 4A-B). From these morphological observations, both carpels and stamens from male and female flowers of L. sisymbrioides, respectively, appeared to abort at intermediate stages of flower development. SEM of flower development in all three species showed flower meristems that initiated from the inflorescence meristem in a similar fashion to Arabidopsis ( Figure 5A-B, D, Arabidopsis stages 1 to 2). As expected, sepal primordia developed first ( Figure 5A-D, stages 3 to 4), followed by petals ( Figure 5B), then presumably stamen and gynoecium primordia. Stamen filaments and anther locules differentiated within the androecium and the gynoecium developed as a tube through postgenital fusion of two carpels ( Figure 5E-F, stages 7 to 8). Subsequently, L. naufragorum started to show more petal expansion than the other two species (compare Figure 5E-H). The gynoecial tube then closed at completion of postgenital fusion and began to differentiate a stigma with papillae ( Figure 5I-J, L, stage 11). In staminate flowers of L. sisymbrioides, after filaments and anther locules of the androecium had differentiated from one another, the gynoecium was arrested in its development ( Figure 5K). The carpels of functional gynoecia expanded laterally, elongating and reaching full maturity with a clearly differentiated style and stigma ( Figure 5M-N, P, stage 12). In L. sisymbrioides staminate flowers, the gynoecium remained aborted at maturity ( Figure 5O) in comparison to functional gynoecia described above. Stamen filaments continued to elongate ( Figure 5M-O, stage 12), except in carpellate flowers of L. sisymbrioides where they remained much shorter than the gynoecium ( Figure 5P). In L. naufragorum, the only species with noticeable petals when mature, petals continued to expand, reaching the length of stamens ( Figure 5M). In the other two species, petal primordia were initiated ( Figure 5F-H) but never expanded, remaining small throughout development ( Figure 5J-L) and not visible at maturity (Figure 5N-P).
Histological sections were performed to further investigate the anatomical development of stamens and carpels. Lepidium naufragorum [see Additional file 2] and L. tenuicaule revealed comparable reproductive organ development with no evidence of loss of organ function. Therefore, only data from L. tenuicaule is compared here against L. sisymbrioides ( Figure 6).
In hermaphroditic flowers of L. tenuicaule, after stamen filaments and anther locules differentiated ( Figure 6A), anthers consisted of PMCs, tapetum and two outer anther This study, estimated based on chromosome counts (Figure 3). c [65]. wall layers (middle layer and endothecium; Figure 6D, J, stage 9) and ovules began to develop in gynoecia ( Figure 6E, K, stage 9). After meiosis of PMCs, anther wall layers degenerated, microspores underwent mitosis, and integuments enclosed the ovule [see Additional file 2, stage 12]. Subsequently, the androecium and gynoecium matured ( Figure 6P). At this stage, the stamen filaments elongated ( Figure 6P) and pollen sacs were composed of a single endothecium layer with secondary wall thickenings ( Figure 6S, Y, stage 13). Pollen grains could be visualized with evident exine and the tapetum had degraded ( Figure 6S, Y, stage 13). By this stage, the gynoecium had fused, elongated, and differentiated a style and stigma ( Figure 6P, stage 13), and ovules had differentiated within each carpel ( Figure 6T, Z1). Apical ovules consisted of an elongated funiculus and an embryo sac, surrounded by the nucellus and two integuments ( Figure 6T, Z1, stage 13).
In staminate flowers of L. sisymbrioides, histological sections revealed that after initiation of the gynoecium ( Figure 6B), sporogenous tissue (PMCs) was present in stamen locules ( Figure 6F, L) and ovules had been initiated ( Figure 6G, M, [see Additional file 2, stage 9]). However at later stages ( Figure 6Q, stage 12; [see Additional file 2, stage 11]), as microspores matured within the anthers and the tapetum degenerated (Figure 6 Z2), the gynoecium failed to elongate and differentiate a style and stigma, and ovules did not grow nor differentiate ( Figure 6U-V, Z3, [see Additional file 2]). Since the gynoecium arrest occurs before microsporogenesis (the production of tetrads from PMCs, stage 9), which normally precedes megasporogenesis (stage 11) in Arabidopsis, we conclude that the process for the loss of gynoecium function in male flowers is the arrest of development at a pre-meiotic, intermediate stage (stage 9).
In young carpellate flowers of L. sisymbrioides ( Figure 6C), sporogenous tissue (PMCs) inside the stamen locules ( Figure 6H, N)  ( Figure 6A, D-E, J-K, stage 9). By the time the gynoecium closed and a stigma and ovule began to differentiate ( Figure 6R, Z5, [see Additional file 2, stage 11]), vacuolated cells pervaded anthers and PMCs had degenerated ( Figure 6W-X, Z4). Stamen filaments did not elongate and neither tetrads, microspores, nor pollen were produced; pollen sacs appeared shrunken, filled with vacuolated cells, and no endothecium layer developed ( Figure 6W-X, Z4). Based on the above observations, we propose that the developmental process for loss of androecium function in female flowers is likely cell death, as evidenced by vacuolated cells (absence of stained cytoplasm) and nuclear degradation ( Figure 6W-X, Z4) following the development of PMCs ( Figure 6H, N). In conclusion, androecium abortion in female flowers occurs at a comparable pre-meiotic stage to gynoecium abortion in male flowers (stage 9) but due to different processes, that is, cell death versus developmental arrest, respectively. Programmed cell death in anther walls is involved in the degradation of pollen mother cells of female L. sisymbrioides flowers Because we were finding evidence of cell death in stamens of female L. sisymbrioides, we wanted to determine whether PCD could be responsible for this abortion of stamens. To look for evidence of PCD, as characterized by DNA fragmentation, we conducted TUNEL assays on histological sections of carpellate L. sisymbrioides and, for comparison, staminate L. sisymbrioides and L. tenuicaule flowers. The TUNEL assay attaches fluorescein to fragmented DNA, eliciting a green fluorescent signal in nuclei undergoing DNA degradation. Lepidium tissue autofluoresced in the absence of staining under both DAPI and fluorescein excitation and emission ranges: cell walls, chloroplasts, nuclei, and pollen grains showed background signal (compare Figure 8A-B to C-D, G-H to I-J, and M-N to O-P). This autofluorescence contributed additional histological evidence that cell death was occurring in stamens of female L. sisymbrioides, as evidenced by the absence or degradation of cell walls, nuclei, and pollen in the center of anther locules, where sporogenous tissue leading to pollen normally develops (compare Figure 8I-J to C-F and O-R). In spite of this autofluorescence, the use of negative and positive controls allowed us to observe strong, above-background, fluorescein signal in certain tissues at certain stages that indicate DNA degradation. For example, all nuclei in the endothecium of mature, functional anther sacs of L. tenuicaule at stage 13 showed a strong, above-background, fluorescein signal indicating PCD (compare Figure 8B to D, red arrow denotes one exemplary nucleus). More importantly, we observed strong, above-background, fluorescein signal in all nuclei throughout all anther wall layers of mature anthers from female L. sisymbrioides (compare Figure 8H to J, red arrows denote exemplary nuclei from each layer). This was taken as evidence that these nuclei are undergoing DNA degradation, as observed in positive controls (treated with DNase) showing higher than abovebackground signal (compare Figure 8B to L, red arrow denotes one exemplary nucleus). When comparing anthers from female L. sisymbrioides that abort at stage 9 to functional anthers from L. tenuicaule and male L. sisymbrioides at the same stage (compare Figure 8I-J to E-F and Q-R), it appeared that PCD in anther wall layers was contributing to the degradation of PMCs in carpellate L. sisymbrioides flowers, in which tapetum, endothecium, and tetrads do not develop as in functional anthers from L. tenuicaule and male L. sisymbrioides at the same stage. In summary, using the TUNEL assay as a proxy for PCD, we find evidence for PCD in the anther wall layers of carpellate L. sisymbrioides flowers, which likely contributes to the degradation of PMCs and abortion of anthers at stage 9.

Discussion
Lepidium sisymbrioides is the sole dioecious member of Brassicaceae, and our phylogenetic analyses show that it is closely related to three other New Zealand hermaphroditic species: L. kirkii, L. naufragorum and L. tenuicaule (Figures 1, 2). Increased phylogenetic sampling of the PI first intron among AANZ taxa allows us to identify reticulation events leading to L. sisymbrioides, which resulted from three past hybridization events ( Figure 2). Of the three close relatives, the shared reticulation history with L. kirkii is a novel finding. Molecular, cytological, and genome size analyses provide evidence that L. sisymbrioides is an allo-octoploid (Figures 1, 2 and 3, Table 1), with 64 chromosomes and an average holoploid genome size (1C-value) of 621 Mbp. By comparing the floral ontogeny of unisexual flowers in L. sisymbrioides to that of its close relatives and to Arabidopsis thaliana, we show that unisexual flowers in this species arose from selective abortion of reproductive organs at a similar floral developmental stage (Figure 7 stage 9) but by different processes in males and females. Differential abortion of the gynoecium in males appears to result from developmental arrest, while in females anther sterility results from programmed cell death (Figure 6, 8).

Evolution of dioecy and unisexual flowers within Brassicaceae
The evolution of dioecy in Brassicaceae occurred only once in the genus Lepidium. We infer that in L. sisymbrioides, dioecy evolved from hermaphroditism via selective abortion of reproductive organs, as in type I flowers [66], as all other members of Lepidium are hermaphroditic. Our floral ontogeny observations confirmed that reproductive organs are initiated and differentially aborted at the same floral developmental stage in male and female flowers of L. sisymbrioides. The timing of abortion corresponds to Arabidopsis stage 9 (Figures 4, 5, 6 and 7), which is broadly considered an 'intermediate' stage of floral development [22], after primordia initiation but before meiosis. Our ontogeny shows that L. sisymbrioides is representative of the majority of angiosperms with type I unisexual flowers that selectively abort reproductive organs at significantly correlated developmental stages between the two sexes [22]. This evidence suggests that similar regulatory switch points underlie male and female developmental pathways as proposed by Diggle et al. [22] and comparable selective forces are at play in the two sexes. However, the developmental stage and process of reproductive organ abortion in unisexual flowers across angiosperms vary widely with different stages and processes occurring at equal frequencies [22]. With regard to the developmental process of organ abortion in L. sisymbrioides, while sporogenous tissue (PMCs) in stamen locules differentiates ( Figure 6C, H, N), it quickly degenerates during development of carpellate flowers and becomes vacuolated with degraded nuclei ( Figure 6R, W-X, Z4). Programmed cell death, which is involved throughout normal flower development [67], is primarily due to endogenous factors and is evidenced by cell death at a predictable time and location during tissue differentiation [68]. During normal flower development, the tapetum degenerates during microgametogenesis via PCD for proper microspore development and differentiation of pollen, [67]. Other studies have shown that premature tapetal degeneration can lead to male sterility [reviewed in 67]. Therefore, because we observe PCD in anther wall layers before microgametogenesis, this premature tapetal degeneration is likely leading to male sterility in L. sisymbrioides females (Figures 6R, W-X, Z4, 8I-J, [see Additional file 2]).
Two types of PCD occur in plants: autolytic and nonautolytic. The former generally occurs during normal plant development, whereas the latter occurs during plantpathogen interactions [69,70]. Moreover, since loss of cell  Figure 7 Model for flower development in Lepidium study species, with proposed timing of reproductive organ abortion in dioecious L. sisymbrioides. The model is informed by our SEM and histological observations (Figures 5, 6) and follows Arabidopsis thaliana stages, noted on the right [20,61,62]. Entire flower shown in stages 1 to 6, only one stamen (left) and one carpel of the bicarpellate gynoecium (right) shown in stages 7 to 13. A filament differentiates within the stamen in stage 8, and elongates in stage 9. ⊢ stage of abortion for female (left) and male (right) flowers of L. sisymbrioides. walls and cytoplasm, nuclear condensation, and increase in vacuolar volume are characteristic of autolytic PCD [70], this type of cell death is also likely involved in the degeneration of PMCs in L. sisymbrioides females ( Figures 6R, 8I-J, [see Additional file 2]).
In male flowers of L. sisymbrioides, on the other hand, the development of ovules and gynoecia is arrested shortly after initiation of ovule primordia. We found no evidence of cell death, parenchymatization, or change in timing of otherwise normal developmental events in arrested gynoecia. Ovule primordia remain evident in mature male flowers ( Figure 6Q, U-V, Z3, Additional file 2). Therefore, of the six developmental processes reviewed in Diggle et al. [22], arrest of development best characterizes the abortion of the gynoecium in L. sisymbrioides males.

Whole genome duplication events via hybridization in the evolution of dioecy in Lepidium
Two different copies of the PI first intron were previously identified among the ANZ taxa (clades A, C); only one copy (clade C) was strongly supported [5]. Our PI phylogeny recovered at least four divergent copies of the first intron in L. sisymbrioides and its close relatives (Figure 1), suggesting ancient allopolyploidization events, followed by divergence of PI alleles. Based on our phylogenetic network analyses, L. sisymbrioides has a history of three allopolyploidization events: hybridization (1) between an Australian and Australian or New Zealand species, (2) with an Asian species, and (3) with an American species (Figure 2). Our study provides new evidence of an additional genome within the ' A' clade of PI [5] and shows an additional PI copy in taxa from this clade (as in, L. kirkii, L. naufragorum, L. sisymbrioides and L. tenuicaule; A1-A2, Figure 1), which would be expected of four divergent genomes contributing to several allopolyploidization events in our study species. Based on our cytological and genome size estimates, L. sisymbrioides is an octoploid, which would require several whole genome duplications. Together with our PI data, this evidence suggests that L. sisymbrioides is an allo-octoploid composed of four different genomes. Australian Lepidium appear to have undergone a rapid radiation during the Pliocene and Pleistocene, when the arid and cooler regions of the southeastern temperate biomes were expanding [71]. Previous studies suggested at least one dispersal event each from California and South Africa to Australia or New Zealand; most likely colonizing Australia first, with at least two subsequent dispersal events to New Zealand [12]. The majority of Lepidium species produce mucilaginous seeds that adhere to birds [4], which may have facilitated long-distance dispersal among the Americas, Australia, New Zealand and the Old World [72][73][74][75]. Our results suggest that a hybridization event occurred either within Australia or between an Australian and a New Zealand ancestor, followed by hybridization with an Asian colonist and an American colonist, resulting in the evolution of L. sisymbrioides (Figures 1, 2). Colonization by an Asian ancestor is not well supported by our data and conflicts with previous studies indicating colonization by an African ancestor [12]; this contradictory evidence could be due to the use of different nuclear DNA regions. In spite of this, our results confirm at least two dispersal events to Australia or New Zealand from the New World and Old World that resulted in allopolyploidization, but exact New and Old World ancestry is uncertain. Additionally, we infer a hybridization between ANZ taxa not previously suggested.
Our 1C-value estimates for L. sisymbrioides and L. tenuicaule fall within the reported range for the family (0.15 to 2.43 pg [76][77][78]). Lepidium naufragorum lies outside the high end of the range, consistent with it being highly polyploid (18x [65]). Even though the 1C-value of L. sisymbrioides (0.635 pg) is almost fourfold that of Arabidopsis thaliana (0.16 pg [79]), it is comparable to the size of other model plants such as rice (0.5 pg [80]), making it a likely candidate for whole genome sequencing. Genomic resources for this species would facilitate the investigation of sex determination and of the putative chromosomal rearrangements that contributed to the evolution of dioecy after polyploidization. As new technologies and approaches are being developed [81,82], sequencing this octoploid will become more feasible in the near future.

Conclusions
The developmental process leading to the evolution of dioecy in Lepidium sisymbrioides was placed in the broader context of the historical patterns conditioning the evolution of separate sexes in this unique dioecious relative of Arabidopsis. We have characterized the developmental stage and process of its unisexual flowers, paving the way for future studies aimed at unraveling the genetic basis underlying reproductive organ abortion. Having placed L. sisymbrioides in a phylogenetic context, determined its ploidy, hybridization history, and genome size, and compared it to Arabidopsis thaliana flower development will facilitate the investigation of the role of polyploidy and of potential candidate genes in the evolution of dioecy in Brassicaceae.