Evolution of floral diversity: genomics, genes and gamma

Abstract

A salient feature of flowering plant diversification is the emergence of a novel suite of floral features coinciding with the origin of the most species-rich lineage, Pentapetalae. Advances in phylogenetics, developmental genetics and genomics, including new analyses presented here, are helping to reconstruct the specific evolutionary steps involved in the evolution of this clade. The enormous floral diversity among Pentapetalae appears to be built on a highly conserved ground plan of five-parted (pentamerous) flowers with whorled phyllotaxis. By contrast, lability in the number and arrangement of component parts of the flower characterize the early-diverging eudicot lineages subtending Pentapetalae. The diversification of Pentapetalae also coincides closely with ancient hexaploidy, referred to as the gamma whole-genome triplication, for which the phylogenetic timing, mechanistic details and molecular evolutionary consequences are as yet not fully resolved. Transcription factors regulating floral development often persist in duplicate or triplicate in gamma-derived genomes, and both individual genes and whole transcriptional programmes exhibit a shift from broadly overlapping to tightly defined expression domains in Pentapetalae flowers. Investigations of these changes associated with the origin of Pentapetalae can lead to a more comprehensive understanding of what is arguably one of the most important evolutionary diversification events within terrestrial plants.

This article is part of the themed issue ‘Evo-devo in the genomics era, and the origins of morphological diversity’.

1. Introduction

The flowering plants (angiosperms) constitute the largest and most diverse extant group of the plant kingdom. Approximately 350 000 species of flowering plants, classified in 416 families and 14 559 genera, have been recorded to date, accounting for nearly 90% of all known land plant species [1,2]. The angiosperms are also the youngest of the major green plant lineages, having arisen and radiated long after plants colonized the terrestrial habitat about 500–470 million years ago (Ma) during the Ordovician ([3]; see also Harrison [4]). A diverse assortment of angiosperms appears abruptly in the fossil record of the Early Cretaceous, starting approximately 125 Ma [5], and representatives of all major extant flowering plant lineages can be recognized in Mid-Cretaceous deposits, about 100 Ma [6]. Molecular-based estimates suggest a somewhat older origin of angiosperms, ranging from 180 to 140 Ma, but, consistent with the fossil record, they support a rapid radiation occurring 5–10 Myr after the evolution of the flowering plant lineage [7–9]. The precipitous origin and rapid diversification of flowering plants was famously referred to as an ‘abominable mystery’ by Charles Darwin because their rapid appearance contradicted his gradualist view of evolutionary change [10]. The ‘mystery’ has duly received considerable attention from developmental and evolutionary plant biologists, with the two major fields of enquiry providing plausible solutions.

The most conspicuous key evolutionary innovation of angiosperms is the flower itself, and breakthroughs in floral developmental genetics provided new impetus for studies of floral evolution and development—floral evo-devo—from which have emerged numerous new hypotheses [11,12]. Among these are novel ideas about how flowers evolved from transformed gymnosperm cones [13–16], an ancestral ‘fading borders’ model for flower development [17–20] and floral diversification through ‘sliding boundaries’ of organ identity functions [21–24]. The prototypical flower is composed of four types of organs arranged such that carpels (the female reproductive organs, collectively the ‘gynoecium’) are innermost and surrounded by stamens (the male reproductive organs, collectively ‘androecium’) which are, in turn, surrounded by petals (usually colourful, collectively ‘corolla’) and then sepals (leaf-like, collectively ‘calyx’). The corolla and calyx collectively constitute the perianth. Variations in the number and arrangement of these four primary floral organs account for much of floral diversity, and can now be understood in the context of genetic specification of floral organ identity [25–27] and floral symmetry [28]. Floral evo-devo studies also offer explanations for the origins of stamens and carpels from gymnosperm precursors via ‘mostly male’, ‘out-of-male’ and ‘out-of-female’ mechanisms [14–16], and the origins of petals from stamens (andropetals) or bracts (bracteopetals) during the course of angiosperm diversification [29,30].

Complementing these developments in floral evo-devo, analyses of the burgeoning collection of flowering plant genome sequences have suggested a role for whole-genome duplications (WGDs; i.e. polyploidy) in the origin and subsequent diversification of flowering plants [31–37]. For example, an ancient polyploidy event has been inferred for the common ancestor of all angiosperms [38,39], three sequential polyploidy events in the monocots pre-date the radiation of the grasses [40,41] and ancient hexaploidy characterizes most eudicots [42–45]. Additional WGDs have been identified among many relatively younger branches of the flowering plant evolutionary tree, mostly among the eudicots [46], many of which coincide closely with the Cretaceous/Tertiary (K/T) boundary about 65 Ma [32]. Moreover, genes involved in signalling and transcriptional regulation tend to be preferentially retained in duplicate following WGD, expanding the repertoire of genetic tools with which evolutionary novelties may be constructed [47–50]. Thus, WGDs and their impact on genes directing floral development and other processes may have been especially important factors in the evolution and diversification of angiosperms [51]. By contrast, WGD may have been less important than tandem gene duplication during animal evolution (see Holland et al. [52]).

Here, we review the current understanding of the evolutionary context from which the most diverse group of extant flowering plants, Pentapetalae, emerged. We emphasize the diverse contributions of phylogenetics, genetics and genomics to understanding key evolutionary changes associated with the Pentapetalae radiation and relate new analyses to unresolved questions surrounding enigmatic WGD events that pre-date their origin.

2. Angiosperm phylogeny: emergence and radiation of the Pentapetalae

Improved understanding of the relationships among flowering plant lineages has provided an ever-expanding framework for hypothesis testing. Amborella, Nymphaeales (water lilies) and Austrobaileyales are successive sisters to the remaining angiosperms (Mesangiospermae), which comprise three major lineages: magnoliids, monocots and eudicots [53] (figure 1). The eudicots are the largest of the extant angiosperm clades accommodating approximately 75% of angiosperm diversity [2]. The eudicot clade arose early in angiosperm evolution, perhaps within approximately 10 Myr of the initial angiosperm radiation [8,9], and is well supported by biochemical (e.g. production of ellagic and gallic acids), morphological (tricolpate or tricolpate-derived pollen) and a wealth of DNA sequence data [53–56]. Inferences of relationships among the eudicots [2,53,55,57] have so far been derived largely from chloroplast molecular sequence data [53,55,57], but given the rapid diversification of eudicots [8,9,58], as well as multiple WGD events [37], these maternally inherited markers may be revealing only a partial glimpse into the evolutionary history of the clade. Current chloroplast-based estimates indicate with strong support that Ranunculales are sister to all other eudicots; Proteales (including Sabiaceae) diverge next; either Trochodendrales or Buxales are successive sister lineages to the rest, although their relative positions remain uncertain; and Gunnerales are sister to a large clade that has been formally named Pentapetalae [53,59]. Pentapetalae alone accommodates about 70% of extant angiosperm species, and together with Gunnerales constitute the Gunneridae (core eudicots). Thus, the eudicots appear to be represented by relatively species-poor lineages (Ranunculales, Proteales, Trochodendrales, Buxales and Gunnerales) that form a basal grade subtending the Pentapetalae, the largest and most diverse group of extant angiosperms.

Figure 1.

Phylogenetic relationships among the main lineages of flowering plants and their sister group, the extant gymnosperms, based on nuclear, mitochondrial and chloroplast DNA sequence data. Amborellales, Nymphaeales and Austrobaileyales form a basal grade below all other flowering plants (Mesangiospermae). Relationships among the three major clades of Mesangiospermae (magnoliids, monocots and eudicots) and Ceratophyllales are currently unresolved. Among eudicots, Ranunculales diverge first, followed by Proteales before a trichotomy comprising Buxales, Trochodendrales and core eudicots. Among core eudicots, Gunnerales are sister to Pentapetalae, which comprise the large Superrosidae and Superasteridae clades and Dilleniales in an unresolved a trichotomy. Superrosidae includes Saxifragales plus Rosidae (rosids), while Superasteridae comprises a basal grade of Santalales, Berberidopsidales and Caryophyllales subtending the large Asteridae (asterids) clade.

Pentapetalae comprises two major clades, formally named Superrosidae (superrosids) and Superasteridae (superasterids), each accommodating approximately one-third of extant angiosperm species [53,60]. Superrosidae and Superasteridae each include a major subclade that corresponds well with morphology-based classifications (e.g. [61,62]), Rosidae (rosids) and Asteridae (asterids), respectively (figure 1). Readily observable floral features that generally distinguish rosids from asterids (figure 2a–c) include: (i) petals free versus fused and (ii) stamens in two whorls and not fused with petals versus a single whorl of stamens fused with petals. The fusion of petals (sympetaly) into a tubular corolla in most asterids has been recognized as a morphological innovation for centuries [63,64]. Other floral features have evolved repeatedly among rosids and asterids, but tend to be more frequent in one or the other. For example, flowers of rosids tend to be small and simply constructed with radial symmetry, while asterid flowers are often elaborate and complex with bilateral symmetry [65].

Figure 2.

Representatives of eudicot floral diversity. (a) Saxifraga rotundifolia (Saxifragales) displays the five-part flowers typical of the Superrosidae clade of Pentapetalae. (b) The five petals are fused in Petunia sp. (Solanales), as is characteristic of the Asteridae clade of Pentapetalae. (c) Antirrhinum majus (snapdragon; Lamiales), a model species for floral symmetry developmental genetics, has zygomorphic flowers in which dorsal, lateral and ventral petal lobes emerge from the corolla tube. (d) Gunnera (Gunnerales) flowers are minute and densely packed in spicate inflorescences. (e) Trochodendron aralioides (Trochodendrales) flowers are polymerous with numerous stamens and carpels. (f) Inflorescences of Pachysandra procumbens (Buxales) bearing dimerous flowers, each with two pairs of stamens. (g) Grevillea sericea (Proteales) flowers displaying four sepal lobes (dimerous) to which stamens are fused, and an elongated pistil. (h) Nelumbo nucifera (Proteales) flowers are polymerous with numerous petals, stamens and carpels. (i) Eschscholzia californica (Ranunculales) flowers are dimerous with two pairs of decussate petals. Photo credits: (a) ‘Saxifraga rotundifolia’ (CC BY-NC 2.0) by cetp; (b) ‘Petunia’ (CC BY-NC-ND 2.0) by Ava Babili; (c) ‘Antirrhinum majus ‘ (CC BY-NC-SA 2.0) by francesco_43; (d) ‘Gunnera’ (CC BY-NC-ND 2.0) by allisoncake; (e) ‘Trochodendron aralioides ‘ (CC BY-NC-SA 2.0) by dogtooth77; (f) ‘Pachysandra procumbens (Allegheny spurge)’ (CC BY-NC-SA 2.0) by tgpotterfield; (g) ‘Grevillea sericea’ (CC BY-NC-SA 2.0) by Marine Explorer; (h) ‘sacred lotus’ (CC BY-NC-SA 2.0) by faria!; (i) ‘Eschscholzia’ (CC BY-NC 2.0) by Nickiz77.

3. Floral roots of Pentapetalae

The most striking feature of Pentapetalae, reflected in the name of the clade, is the transition to a highly conserved, canonical floral ground plan consisting of: (i) whorled arrangement of organs (whorled phyllotaxis); (ii) a fixed merosity or merism (number of organs per whorl); (iii) an ancestrally five-parted (pentamerous) calyx, corolla and androecium (with transitions to four-parted (tetramery) and other merosities); (iv) alternation of organs in adjacent whorls and (v) a single whorl each of sepals and petals [59,65–67]. This canonical floral ground plan (figure 2a) represents a marked departure from the variable arrangement, merosity and morphology of floral organs in early-diverging eudicot lineages [67–69]. Although some of these characters are occasionally found outside Pentapetalae, all five together are a hallmark of this clade. The distribution of floral features among extant basal eudicots suggests that this suite of characters was established along the immediate stem lineage of Pentapetalae, after their divergence from Gunnerales [66,70].

Gunnerales produce dimerous flowers, with two of each type of floral organ (i.e. two sepals, two petals, two stamens and two carpels), all of which may be greatly reduced (figure 2d), perhaps reflecting a shift to wind pollination in this group [70,71]. Flowers of Tetracentron (Trochodendrales) are dimerous and may show intrafloral switches in merosity, changing from dimerous perianth and androecium to tetramerous gynoecium [72,73], and those of Trochodendron (also of Trochodendrales) are polymerous (figure 2e) [65]. The flowers of Buxales (figure 2f) are predominantly dimerous, but with shifts in merosity involving the inner organs of flowers (e.g. tetramerous androecia and trimerous gynoecia are common), and shifts in phyllotaxis that correlate with flower sex (i.e. female flowers are spiral, while male flowers are whorled) [74]. In Proteales, flowers of Proteaceae are dimerous (figure 2g), those of Platanaceae exhibit a shift to trimery and tetramery, those of Nelumbonaceae are polymerous, with a greatly expanded and variable number of floral organs (figure 2h) and those of Sabiaceae represent another independent derivation of pentamery but with spiral phyllotaxis [65,66,75,76]. Among Ranunculales, the sister lineage of all other eudicots, dimerous, trimerous and pentamerous flowers all occur within Ranunculaceae [24,30,66,77]. As in Sabiaceae, and unlike in Pentapetalae, pentamery in Ranunculaceae is coupled with spiral (rather than whorled) organ initiation, and therefore represents another independent derivation of this kind of flower organization [66,77,78] Thus, the non-Pentapetalae eudicots are predominantly dimerous with a few notable exceptions.

4. Genetic origin of the Pentapetalae flower

The whorled pentamerous flower established in Pentapetalae is potentially a key innovation contributing to the success of the clade [79], but the genetic basis of these traits remains unclear. Of the several transcription factor families that are known to play a role in flower development and morphogenesis (e.g. MADS, TCP, MYB, CUC and YABBY), none is a smoking gun for the transition to a whorled pentamerous flower, although they all have potential functions that may have contributed. Together, these gene families pattern the development of morphological traits, such as organ identity, symmetry, fusion, polarity, elongation and growth.

Phylogenetic analyses suggest that all of the MADS genes that regulate floral organ identity experienced either one or two duplication events prior to the radiation of Pentapetalae [80–83]. As a result, Pentapetalae lineages maintain either two or three paralogous forms of each of these genes. Along with the increase in number of MADS genes in Pentapetalae, the spatial expression of these genes shifted from the broadly overlapping ‘fading borders’ pattern of basal angiosperms to sharply restricted expression domains [18,84–86]. Similar evolutionary changes are reported in comparisons of whole transcriptional programmes in floral organs [20,85,87,88]. Much progress has been made in our understanding of floral developmental genetics in Ranunculales [78,89–94], but the basal eudicot grade has not been representatively studied to date. The available data suggest that genetic programmes for floral organ identity are often more broadly deployed in the flowers of non-Pentapetalae angiosperms than in the flowers of Pentapetalae.

Similar to patterns seen in floral MADS-box genes, increased numbers of floral symmetry genes are also associated with the origin of Pentapetalae. Most Pentapetalae flowers are oriented such that there is a single ventral or abaxial petal, two lateral petals and two dorsal or adaxial petals [95]. In radially symmetrical groups, the five petals are identical in form and equidistant from each other, but there have been frequent transitions to bilateral symmetry in which the petals act as three separate modules (dorsal, lateral and ventral). Floral symmetry genes appear to function in these three modules of the flower independently to produce complex petal arrangements—a phenomenon with multiple, independent derivations [96,97]. The primary genetic regulators of floral symmetry are the CYCLOIDEA (CYC) TCP domain transcription factors and the MYB domain transcription factors DIVARICATA (DIV) and RADIALIS (RAD) [28,96]. Phylogenetic analyses suggest that the CYC, DIV and RAD genes expanded into two or three paralogous lineages prior to the origin of Pentapetalae [98–100] and the three CYC clades may have been established through duplications between the divergence of Proteales and the diversification of Gunneridae [101].

A recurrent feature of MADS and TCP genes is poor phylogenetic resolution among triplicated clades that emerged near the origin of Gunneridae. MADS gene trees all show a polytomy below three core eudicot-wide clades [83], as does the gene tree for TCP genes [101]. This lack of phylogenetic resolution may be due to, in part, or in combination with, several factors, including the nature of the duplication event or events (see §5), the rapid speciation of the eudicots and differential gene evolution [101].

5. Origin of the Pentapetalae genome

Arabidopsis thaliana (Brassicaceae; Brassicales), which is the premier genetic model for plant developmental genetics, and Vitis vinifera (grapevine; Vitaceae; Vitales) have been instrumental in shaping our understanding of genome evolution in Pentapetalae. Early examinations of the Arabidopsis genome revealed three WGD events in its evolutionary history, termed alpha (α), beta (β) and gamma (γ) [102,103]. Subsequent analyses of the Vitis genome sequence revealed three large syntenic gene blocks, representing three ancestral genomes brought together in an anciently hexaploid genome (palaeohexaploidy). Importantly, each of the three Vitis syntenic blocks corresponds to four separate regions in the Arabidopsis genome, suggesting that the two WGD events in the Arabidopsis lineage represent the alpha and beta WGDs, while the shared palaeohexaploidy is the gamma event [42]. Each of the Vitis triplicate regions corresponds to two genomic regions in Populus trichocarpa (poplar) [104], reflecting shared palaeohexaploidy followed by a single additional WGD in the poplar lineage. A one-to-one correspondence between Vitis and Carica papaya syntenic regions also indicated shared palaeohexaploidy, but without further WGDs in Carica [105]. All four of these species belong to the rosid subclade of Pentapetalae, but comparisons involving Solanum lycopersicum, Utricularia gibba, Mimulus guttatus and Coffea canephora indicate that the palaeohexaploidy event is shared with these species of the asterid clade and therefore pre-dates the radiation of Pentapetalae [43,106–109]. Notably, like Carica and Vitis, the Coffea genome has not experienced post-gamma WGDs, and as such there exists a 1 : 1 : 1 correspondence between Vitis–Carica–Coffea syntenic regions [109], underscoring their shared palaeohexaploidy.

The triplicate structure of gamma-derived genomes is particularly well preserved in Vitis [42], facilitating intragenomic analyses that explore the historical nature of this hexaploidy. Importantly, two of the three Vitis subgenomes are more fractionated with respect to one another than to the third, suggesting they co-existed in the same nucleus and experienced differential gene loss for a longer period [110,111]. These observations support a ‘two-step’ model for gamma in which the ancestral palaeohexaploid was formed via a tetraploid intermediate in which fractionation was well advanced by the time the third subgenome was added through a wide cross [108,110,111]. Similar fractionation patterns have been found in Brassica rapa, S. lycopersicum and Capsicum annuum, supporting two-step hexaploidies in Brassica and Solanaceae [109,112]. A two-step process for the gamma hexaploidy is also supported by our understanding of the polyploidization process: unreduced gamete formation results in diploid gametes, not triploid ones, and a hexaploid is formed via crossing between a diploid and a tetraploid and further duplication. Thus, hexaploidy is derived via two successive WGDs as in the formation of bread wheat (Triticum aestivum) through a cross between still extant tetraploid and diploid species approximately 8000 years ago [113,114]. However, unlike hexaploid bread wheat, the antiquity of the hypothesized Brassica, Solanaceae and gamma palaeohexaploidies hinders empirical assessment of the two-step hypothesis, and alternative epigenetic modifications could also account for the observed fractionation patterns [112].

Efforts to elucidate the gamma event further have used synteny-based analyses to determine the origin of gamma-derived genomes, and in the absence of genomic data, phylogenomic analyses have been used to estimate the origins of gamma-derived paralogues. It has been established that gamma is absent in Amborella [39], monocots [42], magnoliids, Ranunculales [44] and Proteales [44,45], effectively narrowing the possibilities to the distal branches of the basal eudicots, possibly just prior to the origin of the Gunneridae [83]. The only study implementing both synteny and phylogenomic analyses for a basal eudicot indicated that the triplicate genome structure of gamma does not exist in Nelumbo nucifera (sacred lotus; Proteales), but, surprisingly, approximately 50% of the gene trees support clades that include Nelumbo genes and gamma-derived Vitis paralogues [45]. Close relationships between putative gamma-derived paralogues and basal eudicot genes were found in earlier phylogenomic studies [44,83], but their significance was not explored.

The apparent conflict between synteny and phylogenomics was seen as potentially consistent with the two-step model for gamma palaeohexaploidy [45]. The lack of a gamma-like structure in the Nelumbo genome coupled with phylogenetic grouping of many Nelumbo genes with gamma paralogues could be explained if (i) the initial tetraploidy event in the two-step model post-dates the divergence of Proteales from other eudicots but pre-dates the diversification of Pentapetalae, and (ii) the donor of the third genome to the tetraploid intermediate is a direct, or even older, ancestor of extant Proteales [45]. This speculative two-step scenario was not supported by synteny-based genome halving analyses of the Nelumbo and Vitis genomes [115], which suggested, instead, that the gamma palaeohexaploidy should be placed after the divergence of Nelumbo from other eudicots, somewhere along the stem lineage leading to Pentapetalae. Therefore, to accommodate a two-step model, a more closely related third genome donor than that of a Nelumbo ancestor must be postulated [115]. The three lineages that occupy branches between the divergence of Nelumbo and the radiation of Pentapetalae (i.e. Trochodendrales, Buxales and Gunnerales) are, therefore, pivotal to understanding gamma palaeohexaploidy, but whole-genome sequences for these taxa are not currently available.

6. Towards an elucidation of the gamma event(s)

Previous studies implementing a phylogenomic approach have relied on clustering algorithms, such as OrthoMCL [116], to circumscribe narrowly defined gene families, or orthogroups, the duplication histories of which can be reconstructed phylogenetically [44,45,83]. Such orthogroups ideally define sets of genes descended from a single ancestral gene in the common ancestor of the taxa under consideration [117], but they can be circumscribed more broadly or narrowly depending on the taxon sampling and algorithm settings employed [114]. Therefore, whether putative sets of gamma-derived paralogues are assigned to the same orthogroup, as is necessary for phylogenomic analyses, is a matter of concern that has to be addressed post hoc. For example, only 123 of approximately 1800 gene trees analysed by Vekemans et al. [83] include putative gamma-derived Vitis paralogues, and Jiao et al. [44] combined orthogroups that would have otherwise kept such paralogues separate. Alternatively, in an approach that has not been attempted to date, the synteny-based orthogroups circumscribed for Vitis [43] may be used as a reference to which genes from other species can be assigned, facilitating phylogenomics within the prescribed context of putative gamma-derived paralogues.

Here, we illustrate the use of both the synteny-based orthogroups of Tang et al. [43] and the cluster-based orthogroups circumscribed by OrthoFinder [117] in our own phylogenomic analyses. We include genes from Amborella, two monocots (Oryza and Sorghum), two magnoliids (Liriodendron and Persea) and eudicots representing Ranunculales, Proteales, Trochodendrales, Buxales, Gunnerales and Pentapetalae (table 1). For each orthogroup, protein alignments were constructed using MAFFT [124], converted into nucleotide alignments using PAL2NAL [125], and trimmed by eliminating spurious sequences and alignment positions using trimAl [126]. The resulting orthogroups were then screened for the presence of Amborella (the designated outgroup), Ranunculales, Proteales, duplicate Vitis genes, and at least Buxales, Trochodendrales or Gunnerales using custom Perl scripts. Orthogroups passing these filtering steps were used to construct gene trees with bootstrap support (BS) values (100 replicates) using RAxML [127]. In order to use the resulting trees to trace duplication events as in a previously described pipeline [41], a phylogenetic tree for the included species is required. Given uncertainty of relationships for critically important Buxales and Trochodendrales, we used the MarkerMiner pipeline [128] to construct phylogenetic data-sets based on single-copy nuclear (SCN) loci. Individual datasets were analysed using RAxML as described above to generate species trees using the ASTRAL coalescent approach [129] as well as a supermatrix of the SCN loci (produced using FASconCAT [130]).

Table 1.

Source of datasets for the 30 species included in this study. 1KP, 1000 Green Plant Transcriptome Project [118]; AAGP, Ancestral Angiosperm Genome Project [119], Lotus-DB [120,121], Phytozome [122,123].

species	lineage	source	no. genes/unigenes
Akebia trifoliata	Ranunculales	1KPa	20 366
Amborella trichopoda	basal angiosperm	Amborella genome projectb	26 846
Aquilegia coerulea	Ranunculales	Phytozomeb	24 823
Arabidopsis thaliana	Rosidae	Phytozomeb	27 416
Buxus sempervirens	Buxales	1KPa	20 186
Carica papaya	Rosidae	Phytozomeb	27 751
Citrus sinensis	Rosidae	Phytozomeb	25 379
Eschscholzia californica	Ranunculales	1KPa	26 317
Euptelea pleiosperma	Ranunculales	1KPa	21 659
Glycine max	Rosidae	Phytozomeb	56 044
Grevillea robusta	Proteales	1KPa	16 728
Gunnera manicata	core eudicot	1KPa	16 606
Kalanchoe marnieriana	Superrosidae	Phytozomeb	50 461
Liriodendron tulipifera	magnoliid	AAGPa	12 067
Meliosma cuneifolia	Proteales	1KPa	17 784
Meliosma dillenifolia	Proteales	this studya	33 175
Mimulus guttatus	Asteridae	Phytozomeb	28 140
Nandina domestica	Ranunculales	1KPa	17 453
Nelumbo nucifera	Proteales	Lotus-DBb	26 685
Oryza sativa	monocot	Phytozomeb	39 049
Papaver rhoeas	Ranunculales	1KPa	32 741
Papaver somniferum	Ranunculales	1KPa	32 169
Persea americana	magnoliid	AAGPa	19 335
Platanus occidentalis	Proteales	1KPa	22 347
Populus trichocarpa	Rosidae	Phytozomeb	41 335
Sanguinaria canadensis	Ranunculales	1KPa	18 993
Solanum lycopersicum	Asteridae	Phytozomeb	34 727
Sorghum bicolor	monocot	Phytozomeb	33 032
Trochodendron aralioides	Trochodendrales	1KPa	18 636
Vitis vinifera	Superrosidae	Genoscope 8x Releaseb	30 434

^aTranscriptome assembly.

^bGenome annotation.

Our results indicate that Buxales and Trochodendrales are sister taxa collectively sister to the core eudicots, and that the origins of gamma-derived Vitis paralogues (palaeologues) are concentrated along two consecutive stem lineages immediately ‘below’ the Gunneridae (table 2 and figure 3). In the analyses of synteny-based orthogroups, 410 pairs of Vitis palaeologues could be assessed phylogenetically. The origins of 107 (40 with 50% BS or more) were placed along the branch that immediately precedes Gunneridae. The second prominent set of gamma duplications, 102 in total (55 with 50% BS or more), was placed along the branch subtending Buxales, Trochodendrales and Gunneridae (post-Proteales). A noteworthy proportion of Vitis palaeologues (61 in total; 30 with 50% BS or more) was estimated to have originated prior to the radiation of all extant eudicots. Similarly, substantial numbers of ‘core eudicot-wide’ duplications were also found in the phylogenomic analyses of Jiao et al. [44], but as noted above (see [115]), they do not appear to be relevant to gamma palaeohexaploidy. Analyses of Orthofinder-circumscribed groups showed a similar distribution of duplication events (table 2).

Table 2.

Phylogenetic origin of Vitis paralogue duplications inferred from orthogroup phylogenies.

palaeolog origin	synteny-based orthogroups			OrthoFinder-based orthogroups
	BS ≥ 80	BS ≥ 50	BS ≥ 0	BS ≥ 80	BS ≥ 50	BS ≥ 0
Pentapetalae-wide	0	0	1	0	0	1
Gunneridae-wide	8	40	107	6	43	135
post-Proteales	21	55	102	19	36	78
pre-Proteales	0	4	19	1	5	21
Eudicot-wide	12	30	61	11	33	65
pre-magnoliids	2	5	9	3	5	11
pre-monocots	12	14	16	13	13	17

Figure 3.

Evolutionary origin of gamma-derived Vitis paralogues. The branch labels along the backbone of the phylogenetic tree [total(no. with greater than 50% BS)] indicate the number of Vitis paralogues estimated to have arisen along the respective stem linages assuming strict tetraploidy events. Possible phylogenetic origins of the third subgenome via the two-step model of gamma hexaploidy are indicated by 1, 2 and 3. Scenarios in which a tetraploid is crossed with a close relative that branched off the core eudicot stem lineage (position 1) or off an older stem lineage (position 2) can be reconciled with our gene trees if extensive gene loss and/or extinction is invoked. The scenario of a wide cross between the core eudicot tetraploid and a species that branched off the stem lineage ‘below’ Buxales and Trochodendrales (position 3) is less complex evolutionarily.

These findings are consistent with a two-step model for gamma palaeohexaploidy in which a tetraploidy event occurred in the immediate common ancestor of Gunneridae, followed by donation of the third gamma subgenome from among the ancestors of both Buxales and Trochodendrales. Robustly resolved gene trees with representatives of all three gamma-derived subgenomes were not observed in our dataset, perhaps a consequence of extensive fractionation as previously noted [110]. The two-step scenario is, therefore, largely supported by gene trees that include two duplicate gamma-derived gene lineages: Buxus and/or Trochodendron genes are either (1) sister to duplicate core eudicot gene lineages, or (2) sister to one of two duplicate core eudicot gene lineages. Following the logic outlined by Ming et al. (fig. 3 in [45]), gene tree topologies of type (1) are most parsimoniously interpreted as representing a tetraploidy event along the core eudicot stem branch after the divergence of Buxales and Trochodendrales, but are ambiguous with regard to the origin of the third gamma subgenome. They can be reconciled with loss of any one of the three ancestral genes if the third genome was donated from a branch off the stem lineage of core eudicots below the tetraploidy event (position 1 in figure 3) or loss of a gene donated from a branch off the stem lineage below Buxales, Trochodendrales and core eudicots (position 2 in figure 3). Type (2) gene trees effectively pair a gene lineage that was inherited by core eudicots, Buxales and Trochodendrales, with a core eudicot-specific gene lineage. If the core eudicot-specific gene lineage is one of the tetraploidy-derived duplicates, this topology can be reconciled with organismal phylogeny by postulating a wide cross involving the core eudicot tetraploid and a species that diverged below the common ancestor of Buxales and Trochodendrales (position 3 in figure 3). Alternatively, it is consistent with the donation of the eudicot-specific gene lineage from an extinct line that branched below Buxales, Trochodendrales and core eudicots (at position 2 in figure 3). Thus, barring complex extinction scenarios, our gene trees do not support placing all gamma-associated duplications after the divergence of Buxales and Trochodendrales, nor do they support a WGD in the common ancestor of Buxales, Trochodendrales and core eudicots. Instead, the inclusion of Buxus and/or Trochodendron genes in one of the putative gamma-derived gene lineages is more easily reconciled with the donation of a third subgenome through a wide cross involving an ancestor of Buxales and Trochodendrales, as envisioned in the two-step model (figure 3).

7. Implications of gamma palaeohexaploidy

The close phylogenetic coincidence of the gamma palaeohexaploidy and the origin of pentamerous flowers suggests a causal relationship. As noted, gamma likely arose via a two-step process, with each WGD yielding a set of duplicated genes at each locus. Thus, barring extensive gene loss, we expect a minimum of two or three paralogues for all genes relative to the gene complement present in basal eudicots, monocots and basal angiosperms. In fact, as reviewed in §6, such paralogue diversity is indeed present for many of the key regulators of floral development within Pentapetalae. Especially relevant are transcription factors of the MADS-box, TCP domain and MYB domain gene families, all of which show duplications or triplications prior to the origin of Pentapetalae. For example, multiple duplications in the MADS-box family trace to gamma, and the resulting paralogues of the APETALA1, APETALA3, AGAMOUS and SEPALLATA subfamilies have typically diverged in sequence, expression and function (see [51] for review). Likewise, multiple duplications of TCP genes are also coincident with gamma [101], and paralogous gene lineages have assumed roles in floral symmetry, regulation of vegetative branching and unknown functions in flowers [131–134].

WGD provides the stimulus and genetic raw material for evolutionary novelty [11,37,135]. Although evidence for a causal role of gene duplication in morphological novelty remains limited, data are beginning to accumulate in support of a functional link. For example, differential expression patterns of three paralogues of AP3 coupled with PI control petaloidy in Aquilegia (Ranunculales) and appear to be responsible for the novel features of columbine flowers [90,136]. Duplications of entire genomes allow more complex intergenic interactions, involving multiple paralogues of all genes in the genome, with potentially greater morphological effect than duplications of single genes. Moreover, sequential WGDs, such as those responsible for the palaeohexaploidy recognized as gamma, have even greater potential for novelty than a single WGD.

Narrowing the phylogenetic placement of gamma provides the framework for much more detailed examination of the key features of Pentapetalae. Although we have emphasized the pentamerous, whorled flower of Pentapetalae, other complex floral features, such as bilateral symmetry and highly synorganized flowers (with closely associated floral organs, arising through either fusion or special physical placement of floral parts) also originated within Pentapetalae, perhaps built on the genetic diversity residing in these gamma-derived genomes. Further, because WGD is a common feature of angiosperm evolution, WGDs that both preceded and followed gamma may also have contributed to floral diversity in Pentapetalae. The effects of ancient WGD may not be immediately manifested on a phylogenetic tree; in fact, a phylogenetic ‘lag’ often occurs between WGD and the diversification that may be related to a key innovation [34,137]. Finally, although we focus here on floral traits, we note that other novel features, such as the chemical compound ellagic acid, also trace to Gunneridae or Pentapetalae [49], and further investigation of gamma will have implications for our understanding of many of the key traits that characterize nearly 75% of all angiosperm species.

8. Summary and future prospects

The vast majority of flowering plant diversity can be attributed to the success of a single clade, Pentapetalae, nested within the eudicots. The origin of Pentapetalae coincides with the evolution of a novel suite of floral features (whorled pentamery) and closely follows the gamma genome triplication. These two evolutionary events appear to have had an important impact on flowering plant evolution, but are yet not fully understood. Previous analyses, including the new phylogenomic analysis we present here, have been limited by the available genomic data for three phylogenetically critical lineages: Buxales, Trochodendrales and Gunnerales. These taxa are critical to understanding the timing and nature of gamma palaeohexaploidy, the functional diversification of genes duplicated through this WGD event(s), and relationships between these events and the origin of Pentapetalae. A more fully elucidated evolutionary history of Pentapetalae will, therefore, require the integration of these taxa into several facets of contemporary biological research, including phylogenetics, genomics and functional genetics, which probe the relationships between WGDs, gene duplication, sub- or neofunctionalization, morphological novelty, ecological opportunity and biological radiations.

Data accessibility

Phylogenomic data, including all alignments and trees analysed here, are available from the Dryad Digital Repository http://dx.doi.org/10.5061/dryad.bc80r.

Authors' contributions

A.S.C., B.A.B., D.G.H., D.E.S. and P.S.S. contributed equally to the conception and design of the study. A.S.C. performed the data analyses and drafted the primary manuscript. Additional text and discussion were provided by B.A.B., D.G.H., D.E.S. and P.S.S. B.A.B. and D.G.H. provided data for Meliosma dillenifolia. All authors approved the final version.

Competing interests

We have no competing interests.

Funding

This work was supported by NSF grants no. IOS-1121301, IOS-0922742, DEB-1455601 and DEB-1457440.