Characterization of CYCLOIDEA-like genes in Proteaceae, a basal eudicot family with multiple shifts in floral symmetry

Abstract

Background and Aims The basal eudicot family Proteaceae (approx. 1700 species) shows considerable variation in floral symmetry but has received little attention in studies of evolutionary development at the genetic level. A framework for understanding the shifts in floral symmetry in Proteaceae is provided by reconstructing ancestral states on an upated phylogeny of the family, and homologues of CYCLOIDEA (CYC), a key gene for the control of floral symmetry in both monocots and eudicots, are characterized.

Methods Perianth symmetry transitions were reconstructed on a new species-level tree using parsimony and maximum likelihood. CYC-like genes in 35 species (31 genera) of Proteaceae were sequenced and their phylogeny was reconstructed. Shifts in selection pressure following gene duplication were investigated using nested branch-site models of sequence evolution. Expression patterns of CYC homologues were characterized in three species of Grevillea with different types of floral symmetry.

Key Results Zygomorphy has evolved 10–18 times independently in Proteaceae from actinomorphic ancestors, with at least four reversals to actinomorphy. A single duplication of CYC-like genes occurred prior to the diversification of Proteaceae, with putative loss or divergence of the ProtCYC1 paralogue in more than half of the species sampled. No shifts in selection pressure were detected in the branches subtending the two ProtCYC paralogues. However, the amino acid sequence preceding the TCP domain is strongly divergent in Grevillea ProtCYC1 compared with other species. ProtCYC genes were expressed in developing flowers of both actinomorphic and zygomorphic Grevillea species, with late asymmetric expression in the perianth of the latter.

Conclusion Proteaceae is a remarkable family in terms of the number of transitions in floral symmetry. Furthermore, although CYC-like genes in Grevillea have unusual sequence characteristics, they display patterns of expression that make them good candidates for playing a role in the establishment of floral symmetry.

INTRODUCTION

Basal eudicots form a morphologically disparate paraphyletic grade of 17 families (a total of approx. 4650 species) whose stem lineages diverged prior to the radiation of the core eudicots, the largest group of flowering plants with >250 000 extant species. Floral morphology in these early-diverging eudicot families is diverse and shares some of the characteristics of basal angiosperm flowers, such as variable phyllotaxis (whorled and/or, more rarely, spiral), variable organ number (predominantly dimerous or trimerous) and an undifferentiated perianth (with the exception of Ranunculales) (reviewed in Soltis et al., 2005). Nevertheless, zygomorphy (bilateral floral symmetry), a derived trait generally associated with the fixed floral architectures of monocots and core eudicots (whorled phyllotaxis, trimerous or pentamerous organization), has also evolved independently in five basal eudicot families: Ranunculaceae, Papaveraceae, Menispermaceae, Sabiaceae and Proteaceae (reviewed in Soltis et al., 2005; Damerval and Nadot, 2007; Citerne et al., 2010).

Proteaceae, a moderately sized family (81 genera, >1700 species) of woody plants distributed predominantly in the Southern hemisphere (with hotspots in Australia and South Africa), has received relatively little attention in studies of evolutionary development at the genetic level. Floral organization in this diverse family is remarkably conserved, with four petaloid tepals present in one (undifferentiated) whorl, four antetepalous stamens and a single carpel, yet there is considerable variation in other traits including inflorescence structure, perianth fusion, secondary pollen presentation and floral symmetry (illustrated in Fig. 1). The extensive floral diversification seen within this family might be related to the diversity of pollination strategies (e.g. Johnson et al., 2014), with most species visited by different specific biotic pollinators (insects, birds and mammals) and a few species relying on wind pollination (e.g. Welsford et al., 2014). Inflorescence structure is basically racemose (single flowers, which are derived from racemes, occur rarely across the family), but its individual unit is either a single flower or a flower pair (with or without a common peduncle), the latter being a characteristic of subfamily Grevilleoideae. The axis of bilateral symmetry can differ between flowers of different taxa. In species with racemes or panicles of single flowers, the axis bisects the single adaxial (dorsal) tepal [i.e. flowers have one adaxial (dorsal) tepal, two lateral tepals and one abaxial (ventral) tepal] and the orientation of the flower is said to be anteroposterior. In species with racemes or panicles of flower pairs, six different carpel orientations have been found (Douglas and Tucker, 1996). In three of these, the axis of symmetry of the carpel bisects two opposite tepals; in the other three, the axis of symmetry of the carpel is aligned with gaps between tepals and the orientation of the flower is said to be diagonal. In all taxa with flower pairs, the orientation of each flower is a mirror image of the other in the two-flower unit. Not taking into account carpel symmetry, zygomorphy in Proteaceae can originate from pronounced tepal differentiation, stamen abortion or unequal distribution of hypogynous glands (e.g. Synaphea and certain species of Conospermum), but can also originate from unequal organ fusion (e.g. tribe Proteeae sensu Weston and Barker, 2006) or perianth curvature (e.g. Placospermum). Thirty-four genera of Proteaceae (from all major subfamilies) out of 81 have species with zygomorphic flowers (Weston and Barker, 2006; Reyes et al., 2015). These include some of the largest genera such as Persoonia (100 spp.), Conospermum (53 spp.), Synaphea (50 spp.), Serruria (51 spp.), Protea (114 spp.), Banksia (169 spp.), Grevillea (362 spp.) and Hakea (149 spp.). In addition, some genera are polymorphic, with both actinomorphic and zygomorphic species (e.g. Persoonia, Grevillea and Lambertia) (Reyes et al., 2015).

Fig. 1.

Maximum likelihood reconstruction (equal rates model) of perianth symmetry evolution in Proteaceae on a new species-level tree (best-scoring RAxML tree dated with penalized likelihood). Branches are coloured according to the state with the highest marginal likelihood of their terminal node (or the tip state for terminal branches). Grey branches correspond to missing or inapplicable data (i.e. perianth absent). For full details, see Supplementary Data Figs S1 and S2. The photographs on the right-hand side illustrate some of the variation in perianth symmetry found across Proteaceae. (A) Persoonia amaliae (Persoonioideae), actinomorphic flower. (B) Synaphea sp. (Proteoideae: Conospermeae), zygomorphic flower. (C) Buckinghamia celsissima (Grevilleoideae: Embothrieae), pair of zygomorphic flowers. (D) Virotia leptophylla (Grevilleoideae: Macadamieae), pair of actinomorphic flowers. (E) Banksia ericifolia (Grevilleoideae: Banksieae), pair of zygomorphic flowers. Photograph credits: (A–D) Peter Weston; (E) Hervé Sauquet.

Class II TCP genes from the TEOSINTE BRANCHED1/CYCLOIDEA clade (hereafter referred to as CYC-like genes) have been shown to be crucial for the establishment of bilateral floral symmetry in both monocots and core eudicots, and have been recruited independently in species that have acquired this trait independently (reviewed in Hileman, 2014). Differential expression of these genes has been correlated with independent shifts in floral symmetry in both monocots [e.g. Zingiberaceae (Bartlett and Specht, 2011), Commelinaceae (Preston and Hileman, 2012)] and core eudicots [e.g. Plantaginaceae (Hileman et al., 2003; Preston et al., 2009, 2011; Reardon et al., 2009), Gesneriaceae (Zhou et al., 2008; Song et al., 2009; Yang et al., 2015), other Lamiales (Zhong and Kellogg, 2015), Fabaceae (Citerne et al., 2006), Asteraceae (Chapman et al., 2012), Brassicaceae (Busch et al., 2012), Dipsacales (Howarth et al., 2011) and Malpighiales (Zhang et al., 2010, 2012, 2013)]. The independent recruitment of CYC-like genes for the determination of floral symmetry in both monocots and core eudicots suggests that the ancestral gene from this clade had a predisposition for playing a role in floral development in their common ancestor. A similar role can be expected in basal eudicots, which diverged after the monocot–eudicot split. CYC-like gene evolution in basal eudicots is complex, with frequent clade-specific gene duplications (Citerne et al., 2013). Expression of CYC-like genes has been characterized in Papaveraceae and Ranunculaceae, each of which contains a single transition (or possibly two in the former) from actinomorphy (radial symmetry) to zygomorphy (Damerval and Nadot, 2007; Damerval et al., 2013; Jabbour et al., 2014; Sauquet et al., 2015). However, although CYC-like genes in these families are expressed in flowers and with asymmetric patterns in some species, neither study points to a straightforward role in the control of floral symmetry. In addition, asymmetric expression of CYC-like genes in the only zygomorphic magnoliid Aristolochia was detected only at later stages of floral developmental, after the establishment of perianth zygomorphy (Horn et al., 2015).

Here we combine three complementary approaches to provide a framework for understanding the evolution and genetic control of floral symmetry in Proteaceae. (1) We reconstruct the evolution of floral symmetry (focusing on the perianth) in Proteaceae using a new species-level phylogeny; (2) we characterize CYC-like genes from all major tribes of Proteaceae, and reconstruct their phylogeny; and (3) we compare the expression pattern of CYC-like genes in one actinomorphic and two zygomorphic species of Grevillea, a large and morphologically diverse genus. Combining the results from these three approaches, (1) we did not find any relationship between floral symmetry and the evolutionary history of the CYC-like genes within the Proteaceae; however (2) the asymmetric expression of one CYC-like paralogue in developing zygomorphic flowers of Grevillea suggests that it could be part of the gene network involved in the establishment of floral symmetry in these species.

MATERIALS AND METHODS

Character scoring and ancestral state reconstruction

To reconstruct the evolution of floral symmetry in Proteaceae, we built a new ultrametric species-level tree. Sequence data from the same eight markers used by Sauquet et al. (2009a) were obtained from GenBank for all species of Proteaceae for which data were available (Supplementary Data Table S1). Following initial assessments of single-gene and combined trees, we decided to keep in our sample all species with at least a sequence of matK or of an ITS (internal transcribed spacer), a criterion which allowed us to maximize taxon sampling while avoiding artefacts due to missing data (Wiens and Morrill, 2011). The only exceptions are Beaupreopsis paniculata, Eidothea zoexylocarya and Persoonia falcata, which we decided to maintain in our sample after checking that the other markers sampled for these species allowed a firm placement that was consistent with previous studies. Our final sample includes 213 species, comprising 197 species of Proteaceae and 16 outgroups sampled from all major clades of basal eudicots (Proteales, Sabiaceae, Buxales, Trochodendrales and Ranunculales). In addition, Magnolia grandiflora was sampled in the phylogenetic (RAxML) analysis but pruned out prior to the dating (r8s) analysis. Each of the 81 currently recognized genera of Proteaceae (Weston and Barker, 2006; Mast et al., 2008) is represented by at least one species in our sample. Sampling was independent of the trait considered here (perianth symmetry), except that we ensured that those few genera known to be polymorphic for this trait were represented here by at least one species in each state (Persoonia, Serruria, Grevillea and Hakea). Phylogenetic analyses were conducted with RAxML 8.1.24 (Stamatakis, 2014) on the CIPRES Science Gateway (Miller et al., 2010) using the rapid bootstrapping option, with separate GTRCATI (bootstrap replicates) and GTRGAMMAI (final search) models applied to each marker (partition). The best-scoring RAxML tree was then transformed into an ultrametric tree using the penalized likelihood algorithm of r8s 1.8 (Sanderson, 2003), a smoothing value of 10 (as determined through cross-validation) and the fossil calibrations of Sauquet et al. (2009a).

Floral symmetry was scored in the PROTEUS database (Sauquet, 2015) for the same 213 species as in our phylogeny, following an exemplar approach, using data from various published sources, online floras and examination of preserved and living specimens (Supplementary Data Table S2). Floral symmetry in Proteaceae usually applies to all parts of the flower: the perianth, the androecium and the gynoecium. In addition, the orientation of the single symmetry plane in zygomorphic flowers generally follows carpel orientation. However, there can be exceptions and, thus, for the sake of this study, we recorded here perianth symmetry, regardless of androecium and gynoecium symmetry. We reconstructed ancestral character states for perianth symmetry in Proteaceae using this data set (exported from PROTEUS as a data matrix), the ultrametric tree described above and two approaches: parsimony using Mesquite 3.03 (Maddison and Maddison, 2015) and maximum likelihood (ML) using the rayDISC function of the R package corHMM 1.18 (Beaulieu et al., 2013). For ML reconstructions, we tested the four possible variants for the continuous-time Markov model of discrete binary character evolution (see Sauquet et al., 2015). Here we present the results from the model with the best (lowest) Akaike Information Criterion (AIC) value. We provide the data matrix and ultrametric tree together as a Mesquite-ready NEXUS file (Supplementary Data File S1).

Taxon sampling for CYC-like gene characterization

CYC-like genes were isolated in 33 species from 29 genera (plus Hakea laurina and Leucospermum cordifolium, described in Citerne et al., 2013) belonging to all major tribes of Proteaceae (circumscribed by Weston and Barker, 2006) (Supplementary Data Table S3). Multiple species were sampled from two genera that are polymorphic for floral symmetry: Grevillea (zygomorphic: G. juniperina, G. rosmarinifolia; actinomorphic: G. petrophiloides) and Lambertia (zygomorphic: L. echinata ssp. echinata, L. inermis var. drummondii; actinomorphic: L. formosa). We tried as much as possible to match the taxon sampling for CYC-like gene isolation to a sub-set of that used for our newly reconstructed family-wide phylogeny, but this was not always possible due to the availability of plant material so that different species were sampled for nine genera.

CYC-like gene isolation

Genomic DNA was extracted from fresh or silica dried leaf material following a cetyltrimethylammonium bromide (CTAB) method (Doyle and Doyle, 1987). To determine the number of CYC-like genes, genomic fragments were amplified in 24 species by nested PCR using 1–3 degenerate primer combinations (from Citerne et al., 2013; Supplementary Data Table S4a) designed to bind to conserved regions of the TCP and R domains, and cloned using pGEM-T Easy (Promega) or sequenced directly; sequencing was performed by Genoscreen (Lille, France). Homology to CYC-like genes was determined by the presence of the characteristic TCP and R domains; copy number was determined for each taxon by comparing sequence divergence between clones; clones considered to be of the same type were approx. 99–100 % identical.

In order to characterize the regions upstream and downstream of the conserved TCP and R domains, inverse PCR was carried out in five species (Stirlingia latifolia, Synaphea spinulosa subsp. spinulosa, Protea cynaroides, Triunia youngiana and Banksia integrifolia) representing major tribes within Proteaceae (Ochman et al., 1988; Supplementary Data Table S4b). From the alignment of these sequences, plus the sequences from Leucospermum cordifolium and Hakea laurina (described in Citerne et al., 2013), one and two primers were designed to bind to the region upstream of the TCP and downstream of the R domains, respectively. Fragments were amplified in all species sampled by semi-nested PCR, with the primers above and primers binding to the TCP to R region (Supplementary Data Table S4c), and cloned or sequenced directly as above. Overlapping sequences were assembled with BioEdit v7.2.0 (Hall, 1999). Sequences were deposited in GenBank at the National Centre for Biotechnology Information Database (http://www.ncbi.nlm.nih.gov) (accession numbers are given in Table S3).

Phylogenetic analyses

A total of 55 predicted amino acid sequences from 37 species (including the two published sequences from both Platanus orientalis (Citerne et al., 2013) and Nelumbo nucifera [Citerne et al., 2013; identified within the N. nucifera genome sequence (Ming et al., 2013) using the CoGe (Comparative Genomics) platform: https://genomevolution.org/coge/) included as outgroups] were aligned using MAFFTv.7 (Katoh and Standley, 2013), with manual corrections. Phylogenetic analyses of the nucleotide alignment [a total of 1248 characters, excluding regions of ambiguous alignment (alignment given in Supplementary Data File S2)] were carried out using ML with PhyML 3.0 (Guindon and Gascuel, 2003; Guindon et al., 2010) and Bayesian inference implementing the Markov Chain Monte Carlo (MCMC) algorithm with MrBayes v3.1.2 (Huelsebeck and Ronquist, 2001). The DNA substitution model GTR + I + Γ was specified in PhyML and MrBayes. For the ML analysis, tree search was carried out by Subtree Pruning and Regrafting (SPR) and Nearest-Neighbor Interchange (NNI). Branch support was obtained by approximate likelihood ratio test with the Shimodaira–Hasegawa (SH)-like approach (Guindon et al., 2010) and is given as a percentage. For Bayesian inference, data were partitioned by codon position, allowing variable substitution rates. Two independent analyses (nruns = 2) of four chains (three heated, one cold) were run simultaneously for 2 million generations, sampling every 100 generations. The first 25 % (5000 trees) was discarded as burn-in for each run after checking graphically that convergence had been reached. Majority-rule consensus trees, summarizing topology and branch lengths of the sampled trees and posterior clade probability (PP, given as percentages), were obtained with MrBayes.

Analysis of molecular evolution

Molecular evolution of CYC-like genes in Proteaceae was investigated with codeml from the PAML package v.4.4 (Yang, 2007). In particular, we tested for evidence of positive selection along each paralogous ProtCYC lineage within Proteaceae. The data set consisted of 45 CYC-like genes from 31 Proteaceae species (sequences with missing data included in the gene phylogeny analyses were excluded here) plus the four sequences from Platanus orientalis (Platanaceae) and Nelumbo nucifera (Nelumbonaceae), aligned as described above. The analysed sequence matrix consisted of 273 aligned nucleotides, including the TCP and R domains and excluding regions with gaps in the alignment (Supplementary Data File S3). Initial branch lengths were estimated on the unrooted 75 % majority rule tree inferred by the ML analysis [PhyML 3.0 (Guindon and Gascuel, 2003; Guindon et al., 2010); Supplementary Data Fig. S3] using the one ratio model M0 with the F3 × 4 model of codon evolution. The level of saturation of substitution rates was evaluated with codeml for each branch, and both synonymous and non-synonymous substitution rates were found to be < 1. A range of codon substitution models estimating ω = d_N (non-synonymous rate)/d_S (synonymous rate) across sites [site models M1a, M2a, M3, M7 and M8 (described in Yang, 2007)], and branch-site models estimating an additional ω for a chosen foreground branch were run. Branch-site models either allowed a foreground-specific class ω₂ ≥ 1 (selection model MA), or constrained ω₂ = 1 (null model MA') (Zhang et al., 2005). Nested models were compared by likelihood ratio test against the χ² distribution with degrees of freedom equal to the difference of free parameters.

Gene expression

Fresh flower buds of Grevillea rosmarinifolia and G. juniperina were collected from plants grown in a greenhouse in Gif-sur-Yvette (France), whereas buds of G. petrophiloides were harvested in Sydney (Australia) by two of us (P.H.W. and H.S.) and stored in RNAlater (Ambion) before processing. Grevillea flowers are borne in flower pairs and, because the relationship of individual flowers to an axis is complex (Douglas and Tucker, 1996), floral orientation (i.e. dorsal and ventral specification) is defined with respect to carpel orientation. In the zygomorphic species G. rosmarinifolia and G. juniperina, the carpel axis bisects the two dorsal tepals and stamens and the two ventral tepals and stamens (stamens are opposite and adnate to the tepals). RNA extractions were carried out at three developmental stages defined by bud length: 1, 1–3 mm; 2, 4–6 mm; and 3, 6–7 mm. At the two later stages, dorsoventral differentiation is clearly visible in the two zygomorphic species. Buds at the oldest stage were dissected as follows: the two large dorsal tepals and stamens; the two small ventral tepals and stamens; and the gynoecium. Total RNA was extracted following a method developed for extracting total RNA from Proteaceae tissues using a 2× CTAB-based extraction buffer and lithium chloride precipitation (Smart and Roden, 2010). Total RNA was treated with DNase (Ambion) following the manufacturer’s instructions. RNA yield was low for G. petrophiloides buds stored in RNAlater, and first stage buds and dissections could not be processed further. cDNA was synthesized with Revertaid reverse transcriptase and random hexamers (Fermentas). Copy-specific primer combinations used were: G. juniperina ProtCYC1 F, 5′-TGACCTTGCCCAACTGCTAT-3′; R, 5′-GATACACGACTGTTGGCACG-3′; ProtCYC2 F, 5′-GCCAGAGACTATGAAGGGATGG-3′; R, 5′-CTGACC CTTGTGTGATCGCA-3′; G. rosmarinifolia Prot CYC1 F, 5′-STCRAACCA YAATTRYAGTGCT-3′; R, 5′-GCARGTGAA TTT CTGAAGTAG-3′; ProtCYC2 F, 5′-STCRAACCA YAA TTRYAGTGCT-3′; R, 5′-GCARGTGAATT TCTGAAGTAG-3′; G. petrophiloides ProtCYC1 F, 5′-TCTGAAGC GGAAA TG AGCCC-3′; R, 5′-GGGGTGCTGGGTACTAATCC-3′; and ProtCYC2 F, 5′-TCAAGCCCTTCTGAAACTGGG-3′; R, 5′-TATTGCTGCAAACTGACCCCT-3′. Actin was used as a positive control and quantified on a 1·5 % agarose gel stained with ethidium bromide after 28 cycles (in the exponential phase of amplification); amplified CYC-like gene cDNA was visualized after 35 cycles. At least three biological replicates with 1–3 experimental replicates for each were carried out for G. juniperina and G. rosmarinifolia. For G. petrophiloides, only one biological replicate was available for stage 2, and low yields meant material was sufficient for only one experimental replicate; two biological replicates were replicated twice for stage 3. Amplified fragment identity was verified by direct sequencing (Genoscreen). Transcription of the region upstream of the TCP with stop codons (ProtCYC1) was determined by amplification and direct sequencing using the following primers: G. rosmarinifolia F, 5′-CAACACCGATTTACCTCTGACG-3′; R, 5′-TTCTTCTTAGTGGAGATCGCCT-3′; G. petrophiloides F, 5′-GCTGATGAATGATGAACAACACC-3′; R, 5′-GCCA TCT A  GAAGCAGTTGGG-3′; and G. juniperina F, 5′-CAAC TGC ACTCCATTATATGCTG-3′; R, 5′-CTCAACTAC AACC GCCACTG-3′. As a control for genomic DNA contamination, PCR for actin was carried out for all samples without reverse transcriptase. Quantitative reverse transcription–PCRs (qRT–PCRs) were conducted on the three biological replicates of dissected flower buds and stage 1 flower buds of G. juniperina with the Biorad CFX384 touch and the SYBR Premix Ex Taq (Takara). The efficiency of primer pairs was measured and validated using serial dilutions of cDNAs. We used actin, tubulin and fructose-bisphosphate aldolase as reference genes. The latter two sequences were obtained from a floral G. juniperina transcriptome (E. Delannoy, unpubl. res.; primers are listed in Supplementary Data Table S4d). Three technical replicates were done for each biological replicate. The expression of CYC-like genes in dissected floral organs was calculated relative to stage 1 flower bud and normalized using the arithmetic mean of the relative abundance of the three reference genes. Standard errors were calculated from biological replicates. Gene expression levels were compared between samples using a t-test.

RESULTS

Ancestral state reconstruction

The general phylogenetic backbone of Proteaceae (i.e. relationships among genera) was resolved with very good support (bootstrap values >95 %) overall (Fig. S1). The exceptions are relationships within the tribe Persoonieae, among the main clades of subfamily Proteoideae, within the Leucadendreae, within the Hakeineae, among the four tribes of Grevilleoideae, among the main clades of Macadamieae, within the Gevuininae and within the Roupaleae. Most of the 81 genera are monophyletic, except Persoonia, Sorocephalus, Spatalla, Mimetes, Stenocarpus, Grevillea and Heliciopsis. Relationships within genera are generally poorly resolved. Parsimony and ML ancestral state reconstructions yielded very similar results (Fig. 1; Fig. S2; parsimony results not shown). The equal rates (ER) model was selected as best fit. Although we cannot reject the unequal rates (ARD) model based on its AIC value, the two transition rates estimated with this model were very similar (q01 = 0·0056; q10 = 0·0042) and the reconstructed ancestral states (i.e. states with highest marginal likelihood) were identical for all nodes (ARD results not shown). Conversely, both unidirectional models were strongly rejected (Fig. S2). Our results reconstruct actinomorphy as ancestral in Proteaceae (P = 0·999) and in each subfamily (Persoonioideae, P = 0·881; Symphionematoideae, P = 0·990; Proteoideae, P = 1·000; Grevilleoideae, P = 0·998). Zygomorphy evolved many times independently in Proteaceae, from ten to 18 times in parsimony reconstructions (where several nodes have equivocal ancestral states; File S1), or 16 times in ML reconstructions (Fig. 1; Fig. S2). Zygomorphy characterizes both small clades or isolated species and genera (e.g. Placospermum and Triunia) and some much larger clades such as Proteeae, Embothrieae and Banksieae, which include three of the largest genera in the family (Protea, Grevillea and Banksia). Furthermore, our results support at least four reversals from zygomorphy to actinomorphy (parsimony, 4–12; ML, 6), including isolated species in large zygomorphic clades (Grevillea curviloba, Hakea drupacea and Hicksbeachia pinnatifolia).

Phylogeny of CYC-like genes in Proteaceae

Phylogenetic analyses strongly support the evolutionary scenario of a duplication of a CYC-like gene in the ancestor of all Proteaceae after the divergence of the Proteaceae and Platanaceae lineages (Fig. 2). Two copies, ProtCYC1 and ProtCYC2, were found in species from the major subfamilies Persoonioideae, Proteoideae (tribe Leucadendreae) and Grevilleoideae (tribes Embothreae and Roupaleae). The single copy from Bellendena montana and Symphionema montanum, both basal species in Proteaceae, is nested within the ProtCYC2 clade, suggesting that the ProtCYC duplication pre-dates their divergence. Gene relationships for each paralogue recovered species phylogeny as inferred above and previously. Many taxa sampled here appeared to have only one copy (19 out of 35); the phylogeny suggests that the other copy has been lost or has not been found due to sequence divergence. With the single exception of Protea cynaroides, taxa with only one copy have retained ProtCYC2. There is no obvious correlation between copy number and floral morphology, although in tribe Roupaleae, the representative of the zygomorphic genus Triunia has two copies, unlike the other taxa sampled from this predominantly actinomorphic clade. Apart from Triunia, the two copies were found in clades containing both zygomorphic and actinomorphic taxa.

Fig. 2.

Phylogeny of CYC-like genes in Proteaceae, with CYC-like sequences of Nelumbo nucifera and Platanus orientalis as outgroups. Majority-rule consensus tree from the PHYML analysis of the aligned nucleotide sequences (1248 characters). Clade support is given with Shimodaira–Hasegawa (SH)-like values (PHYML) and posterior probabilities (MrBayes), respectively. Φ denotes zygomorphic species.

Molecular evolution

Models of molecular evolution that allow the non-synonymous to synonymous substitution rate ratio ω to vary along codon sites did not identify any codon under positive selection in Proteaceae CYC-like genes (Supplementary Data Table S5A). Most sites in this analysis were under strong purifying selection (under the M3 model with three site classes, ω₀ = 0·03 for 64 % of sites and ω₁ = 0·31 for 33 % of sites). Branch-site models, which have more power to detect positive selection, were evaluated along the two branches from the CYC-like gene duplication in Proteaceae (Supplementary Data Fig. S3). No evidence of positive selection was detected in predicted coding sequences for either ProtCYC lineage (Supplementary Data Table S5B). However, since only putative coding sequence and aligned regions where no gaps were inserted were considered, a large section of the putative protein was not taken into account in these analyses.

In certain Embothrieae, ProtCYC1 sequences have single nucleotide indels (e.g. Alloxylon pinnatum) and in frame stop codons (Hakea laurina, Grevillea spp.) upstream of the TCP domain. In addition, a unique 36 bp insert (with a string of 14 A nucleotides) was found in this region in G. petrophiloides (Fig. 3). Amplification and sequencing of this region from G. rosmarinifolia, G. juniperina and G. petrophiloides mRNA suggest that this region is transcribed; it is not an intron and thus should constitute a 5'-untranslated region (UTR). The methionine codon allowing the typical TCP and R domains to be translated in these species is not conserved across ProtCYC1 genes and, conversely, the furthest methionine codon predicted to be at the start of the ProtCYC1 open reading frame (from inverse PCR data) is not in-frame with the TCP domain in H. laurina ProtCYC1, reinforcing the idea that the N-terminus of ProtCYC1 proteins may be strongly divergent. In addition, in Proteoideae (Leucospermum cordifolium, Mimetes cucullatus, Spatalla mollis and Sorocephalus lanatus), ProtCYC1 sequences have putatively premature stop codons (approx. 350 bp after the R domain) compared with all other ProtCYC sequences. Comparison of sequence divergence between ProtCYC1 and ProtCYC2 showed that there was little difference in sequence similarity between sequence pairs in regions spanning the TCP domain and a conserved motif in both copies downstream of the R domain (FENWDID) (93·1–98·6 % and 93·9–97·5 % sequence identity for ProtCYC1 and ProtCYC2, respectively, between sequences from Grevillea and Hakea; 89·3–97·9 % and 91–98·8 % sequence identity for ProtCYC1 and ProtCYC2, respectively, between sequences from sampled Leucadendreae). However, the putatively non-coding region upstream of the TCP domain of ProtCYC1was more divergent between Grevillea/Hakea (71·2–90·6 % sequence identity) than the orthologous putatively coding region of sampled Leucadendreae (83·1–97·4 % sequence identity) (Supplementary Data Table S6).

Fig. 3.

Alignment of ProtCYC1 sequences from six species of Embothrieae. Stop codons (*) were found upstream of the TCP domain in Grevillea spp. and Hakea laurina; the alignment suggests a single nucleotide insertion in the sequence of Alloxylon pinnatum. The putative start of the open reading frame (arrows) may be different in these species compared to other sequenced ProtCYC genes such as Telopea mongaensis and Stenocarpus salignus ProtCYC1.

Gene expression in Grevillea species

Flowers of Grevillea are grouped in racemose conflorescences (where flowers are borne in lateral pairs); in G. rosmarinifolia and G. juniperina the axis of symmetry bisects the flower between the two dorsal tepals/stamens and the two ventral tepals/stamens. Flowers of G. rosmarinifolia are strongly zygomorphic, with reduced hooked ventral tepals and stamen filaments; flowers of G. juniperina are also zygomorphic but less bent than in G. rosmarinifolia; flowers of G. petrophiloides have an actinomorphic perianth (Fig. 4). Both ProtCYC copies were expressed in floral buds in all three species and with different expression patterns (Fig. 4). In G. rosmarinifolia, ProtCYC2 expression was detected only in young floral buds, but not at later stages, whereas ProtCYC1 expression was maintained during later stages of floral development. At this later stage, ProtCYC1 was expressed predominantly in the ventral region of the flower, in the tepals and/or androecium. Weak expression was detected for ProtCYC2 in the gynoecium. In G. juniperina, expression of ProtCYC1 appeared greater than that of ProtCYC2 at the later stages, except in the gynoecium. qRT–PCR results appear to show a statistically significant asymmetric expression of ProtCYC1 (P = 0·0168) but not ProtCYC2 (P = 0·1576) between the dorsal and ventral organs of G. juniperina (Fig. 5; Supplementary Data Table S7). In entire floral buds in G. petrophiloides, ProtCYC2 expression was detected only at a later developmental stage, where it is more expressed than ProtCYC1.

Fig. 4.

Expression patterns of ProtCYC1 and ProtCYC2 in developing flowers of three species of Grevillea (zygomorphic, G. rosmarinifolia, G. juniperina; actinomorphic, G. petrophiloides). Floral bud length: 1, 1–3 mm; 2, 4–6 mm; and 3, 6–7 mm. Dissections were carried out on 6–7 mm buds. Do, dorsal tepals and stamens; Ve, ventral tepals and stamens; G, gynoecium. Actin is shown as a positive control.

Fig. 5.

qRT–PCR expression of ProtCYC1 and ProtCYC2 in dissected floral organs of Grevillea juniperina. Dorsal and ventral refer to the tepals and adnate stamens. Expression levels are relative to stage 1 flower bud expression. Error bars represent the s.d. from the three biological replicates. The expression level of ProtCYC1 in the ventral tepals and stamens is significantly greater than that of the dorsal tepals and stamens (*5 % significance level).

DISCUSSION

Evolution of floral symmetry in Proteaceae

Our new species-level phylogeny of Proteaceae is generally consistent with the previous family-wide study (Sauquet et al., 2009a) and recent work on focused clades within the family (Mast et al., 2008, 2012; Sauquet et al., 2009b; Cardillo and Pratt, 2013). Because our main focus here was not to provide a thorough test of these relationships, we did not obtain new sequence data for this study, and many gaps remain in our eight-marker data set. As a result, some relationships are not supported, especially within the larger genera; therefore, we urge caution with use of this tree for purposes other than this study. Although it is possible to take phylogenetic uncertainty into account in ancestral state reconstructions (Pagel et al., 2004), our main objective here was to highlight the pattern of many replicated origins of zygomorphy in Proteaceae, for which we believe the single tree presented here is sufficient. Divergence time estimates obtained here are very similar to those from the PL analysis of Sauquet et al. (2009a). Our new reconstruction of the evolution of perianth symmetry in Proteaceae is consistent with the analyses of Reyes et al. (2015), which were based on the genus-level tree of Sauquet et al. (2009a). However, although this former study highlighted 11 separate origins of zygomorphy and a single reversal, our new results here based on a more comprehensive species-level tree allow us to dissect the evolution of this trait more finely and highlight at least ten (up to 18) separate origins of zygomorphy and four (up to 12) reversals (Fig. 1; Fig. S2). A surprising difference is the lack of support for a strong asymmetry in transition rates in this study, in contrast to the results of Reyes et al. (2015; reversal rate much lower than origin rate). We attribute this difference to taxon sampling, but we note that transition rates can be difficult to estimate accurately and that this question will require further investigation. Likewise, we anticipate that future similar studies with increased species sampling may reveal additional origins (from isolated species) and more reversals. It has long been understood that zygomorphy evolved many times independently in angiosperms (e.g. Citerne et al., 2010). However, to our knowledge, there is no other family in angiosperms that experienced as many within-family origins as Proteaceae did (Reyes et al., 2016).

Evolutionary history of CYC-like genes in Proteaceae

Our phylogenetic analysis of CYC-like genes in Proteales, which included representatives of the basalmost clades of Proteaceae [i.e. Bellendenoideae and Persoonioideae (Weston and Barker, 2006)], showed that a single duplication of these genes occurred after the divergence of Platanaceae and Proteaceae and preceded the diversification of the family. Optimization of floral symmetry character states on the current phylogenetic tree of Proteaceae indicates that the ancestral state for floral symmetry in this family is actinomorphy (Fig. 1). Therefore, neither the duplication of CYC-like genes nor a subsequent loss of one paralogue coincided with a shift in floral symmetry in Proteaceae. As discussed above, zygomorphy is estimated to have evolved at least ten times independently within Proteaceae, with at least four instances of actinomorphy derived from a zygomorphic ancestor (Fig. 1). Several studies have linked duplication or gene loss events with shifts in floral symmetry. For instance, additional duplications were associated with a shift from actinomorphy to zygomorphy in Caprifoliaceae subfamily Dipsacoideae (Howarth and Donoghue, 2005; Carlson et al., 2011), and Ranunculaceae (Jabbour et al., 2014). Gene loss has been associated with a shift from zygomorphy to actinomorphy in Plantaginaceae (Preston et al., 2011). However, no additional duplications or specific gene loss of CYC-like genes were inferred within Proteaceae, and there appears to be no link between floral symmetry type and copy number in this family (Fig. 2). In particular, only ProtCYC2 was isolated in Conospermum longifolium and Synaphea spinulosa, two species belonging to genera with the most pronounced zygomorphic floral development in Proteaceae (strong tepal differentiation and stamen abortion). Less than half of the Proteaceae species sampled, distributed across three different tribes with both zygomorphic and actinomorphic species, appear to have retained both copies.

There appears to be a strong trend to retain ProtCYC2 over ProtCYC1, yet no difference in selective pressures was detected in the main coding regions between the two paralogues. However, differences in evolutionary patterns between the two copies may exist, as suggested by the frequent loss (or divergence leading to the absence of PCR amplification) of the ProtCYC1 gene and specific features of this same gene in certain species. The presence of a number of stop codons upstream of and in-frame with the TCP and R domains, and premature stop codons downstream of the R domain, should result in a shorter N-terminus (in some Embothrieae) or C-terminus (in some Leucadendreae) of the putative ProtCYC1 protein, and may impact its function. In addition, the transcription of the 5′ region containing stop codons in both Grevillea rosmarinifolia and G. juniperina suggests that it is a 5′-UTR rather than an intron that may have specific regulatory functions. To our knowledge, this is the first report of such features in the region upstream of the TCP domain of CYC-like genes.

Possible role of CYC-like genes in floral development

Even though the evolutionary history of CYC-like genes in Proteaceae is distinct from that of core eudicots or monocots, these genes are florally expressed in the three species of Grevillea we sampled, whether zygomorphic or actinomorphic, in a way that is similar to what has been generally found so far in other studies of CYC-like genes (reviewed in Hileman, 2014). In core eudicots, genes that play a role in the control of floral symmetry (recruited independently in both rosids and asterids) have so far been found to belong to the CYC2 gene clade (reviewed in Hileman, 2014), which is core eudicot-specific (Howarth and Donoghue, 2006; Citerne et al., 2013). In monocots, CYC homologues (referred to as TB1-like) have their own complex duplication history, but have also been found to be florally expressed, with asymmetric expression patterns correlated with tepal and stamen development in zygomorphic species of Costaceae (Bartlett and Specht, 2011), Commelinaceae (Preston and Hileman, 2012) and Poaceae (Yuan et al., 2009). Furthermore, CYC-like genes are also expressed in developing flowers in the basal eudicot families Papaveraceae and Ranunculaceae, with expression during early floral developmental stages a seemingly conserved feature in both zygomorphic and actinomorphic species (Damerval et al., 2007, 2013; Jabbour et al., 2014). We provide here another example of floral expression in another basal eudicot lineage, and one which is especially remarkable for its numerous shifts in floral symmetry. Floral expression seems to be a generalized feature of CYC-like genes, and has also been described in both actinomorphic and zygomorphic magnoliids (Horn et al., 2015).

Although the expression of ProtCYC genes is largely overlapping, there are differences in expression levels at later stages of floral development that could suggest a role in the unequal development of tepals and/or stamens in zygomorphic Grevillea species. In both Grevillea species, ProtCYC2 is relatively less expressed than ProtCYC1 in the perianth/stamen organs, but more expressed in the gynoecium, suggesting possible sub-functionalization. In both zygomorphic species, expression levels of ProtCYC1 in dissected flowers were greater in the short ventral tepals/stamens than in the long dorsal tepals/stamens. As in core eudicots, increased expression is associated with reduced organ size. It is worth underlining that asymmetric expression of CYC-like genes in itself, especially if it does not precede the establishment of the phenotype, is not sufficient to establish that these genes play a role in floral symmetry. This was the case for a basal eudicot taxon from the Fumarioideae, where asymmetric expression was only visible after morphological differentiation along the symmetry axis (Damerval et al., 2013). Similarly, in developing flowers of the zygomorphic magnoliid Aristolochia, a predominantly abaxial expression was detected only late in development and did not coincide with the establishment of zygomorphy (Horn et al., 2015). Such late asymmetric expression patterns strongly suggest that CYC-like genes respond to other more precociously asymmetrically expressed genes or signals that are the actual triggering factor(s) of zygomorphy.

CONCLUSIONS

The Proteaceae is one of the largest and most diverse families of basal eudicots. Although the floral groundplan is conserved, there is a remarkable variation in floral morphology and inflorescence structure. Additionally, Proteaceae appears to be a real hotspot for repeated ‘switches’ between the two main types of symmetry and therefore represents a very interesting clade to dissect the evolution of this trait further at the morphological, genetic and functional levels. This study is the first to investigate the evolution and expression of putative floral development genes in this family. Our expression results suggest that CYC-like genes are possible candidates for the control of floral symmetry and will warrant further investigation. The lack of correlation between floral symmetry switches and copy number of CYC-like genes would suggest that gene duplication is not a necessary feature of this morphological change. However, the role of other candidates, such as B-class MADS box genes known to be implicated in the control of floral symmetry in certain monocot lineages (reviewed in Hileman, 2014), should also be examined. It will be important to try to overcome the technical difficulties encountered when working on members of this family to describe the early floral expression pattern of these genes (and test whether they coincide with the establishment of zygomorphy). There is also scope for comparative studies in a larger sample of species in order to decipher the possible genetic changes associated with the multiple transitions in floral symmetry in Proteaceae.

SUPPLEMENTARY DATA

Supplementary data are available online at www.aob.oxfordjournals.org and consist of the following. Table S1: list of GenBank accessions used in the phylogenetic analyses. Table S2: extraction of the PROTEUS database with complete data records and references used to assemble the perianth symmetry data set. Table S3: list of species sampled for the characterization of CYC-like genes and sequence accession numbers of the latter. Table S4: list of primers and primer combinations. Table S5: (A) Parameters of the site models and summary of likelihood ratio tests of nested models of sequence evolution of CYC-like genes from Proteales. (B) Parameters of the branch-site model MA for the ProtCYC1 and ProtCYC2 lineages. Table S6: percentage nucleotide sequence identity between aligned portions of ProtCYC1 and ProtCYC2 in selected Embothrieae and Leucadendreae. Table S7: pair-wise t-test results of comparisons of ProtCYC1 and ProtCYC2 mRNA expression between floral bud stage 1, and dissected floral organs in Grevillea juniperina. Figure S1: best-scoring tree from the RAxML maximum likelihood analysis, annotated with bootstrap support values. Figure S2: full details of the maximum likelihood reconstruction of perianth symmetry evolution in Proteaceae summarized in Fig. 1. Figure S3: phylogeny of Proteaceae CYC-like genes used for inference of molecular evolution by PAML. Data File S1: Mesquite-ready NEXUS file containing the perianth symmetry data matrix and the ultrametric (penalized likelihood) tree used in all analyses of character evolution. Data File S2: edited MAFFT-aligned nucleotide sequences of Proteaceae CYC-like genes analysed for phylogeny reconstruction. Data File S3: aligned nucleotide data matrix of Proteales CYC-like genes used for PAML analyses of codon evolution models.