•The MADS box transcription factor family has long been identified as an important contributor to the control of floral development. It is often hypothesized that the evolution of floral development across angiosperms and within specific lineages may occur as a result of duplication, functional diversification, and changes in regulation of MADS box genes. Here we examine the role of GLOBOSA (GLO)-like genes, members of the B-class MADS box gene lineage, in the evolution of floral development within the monocot order Zingiberales.
•We assessed changes in perianth and stamen whorl morphology in a phylogenetic framework. We identified GLO homologs (ZinGLO1-4) from 50 Zingiberales species and investigated the evolution of this gene lineage. Expression of two GLO homologs was assessed in Costus spicatus and Musa basjoo.
•Based on the phylogenetic data and expression results, we propose several family-specific losses and gains of GLO homologs that appear to be associated with key morphological changes. The GLO-like gene lineage has diversified concomitant with the evolution of the dimorphic perianth and the staminodial labellum.
•Duplications and expression divergence within the GLO-like gene lineage may have played a role in floral diversification in the Zingiberales.
The monocot order Zingiberales comprises a major component of both tropical and subtropical ecosystems and includes crop plants (e.g. banana, plantain and ginger), sources of traditional medicines and spices (e.g. cardamom, turmeric and galanga) and horticulturally important ornamentals (e.g. Heliconia, bird-of-paradise, prayer plants and Canna). The order contains c. 2500 species, many of which form specialized pollination relationships with bees, birds, bats, dung beetles, moths, butterflies and primates (lemurs) via alterations in floral form (Frost & Frost, 1981; Itino et al., 1991; Kress et al., 1994; Sakai & Inoue, 1999; Sakai et al., 1999). The order has long been recognized as a ‘natural’ group of plants (Bentham & Hooker, 1883) and more recent phylogenetic analyses confirm that the Zingiberales forms a monophyletic lineage (Duvall et al., 1993) and is part of the commelinid monocots (APG III, 2009;Davis et al., 2006; Chase et al., 2006). The eight families currently recognized within the order are often divided into two informal groups: the four ‘banana families’ Musaceae, Strelitziaceae, Lowiaceae and Heliconiaceae; and the four ‘ginger families’ Cannaceae, Marantaceae, Zingiberaceae and Costaceae (Kress, 1990a; Kress et al., 2001). A summary of currently accepted phylogenetic relationships within the order is presented in Fig. 1(a).
Several significant changes in floral morphology have occurred through the course of Zingiberales evolution involving the perianth and androecial whorls (Kress, 1990a; Rudall & Bateman, 2004). Of particular relevance to the evolution of pollination syndromes are the derivation of a well-differentiated perianth, the development of petaloid staminodes, and the fusion of staminodes to form the staminodial labellum of Costaceae and Zingiberaceae (Fig. 1). Flowers in the Zingiberales have two trimerous whorls of tepals, two trimerous androecial whorls, and a tricarpellate gynoecium. In the banana families, flowers typically have five or (rarely) six fertile stamens at maturity. In those taxa with five fertile stamens, the sixth androecial member may abort and be completely absent in the mature flower (Strelitziaceae and Lowiaceae, and some Musaceae) or may develop as an infertile staminode (Heliconiaceae and some Musaceae). In the ginger families, the number of fertile stamens is reduced to one (Zingiberaceae and Costaceae) or one-half, with only a single theca (Cannaceae and Marantaceae) (Kirchoff, 2003; Rudall & Bateman, 2004). The remaining androecial members in the ginger families develop as petaloid staminodes. These infertile stamens share positional homology with stamens in the banana families and other monocots, but develop as petaloid structures, taking on the function (pollinator attraction) and structure (conical epidermal cells; C. D. Specht, unpublished) of petals (Kirchoff, 1991). In Zingiberaceae and Costaceae, these staminodes fuse in various combinations to form a novel structure, the staminodial labellum (Kirchoff, 1988a, 1998, 2003). The staminodial labellum provides the majority of the visual floral display and underlies the variety of pollination syndromes found in these diverse families (Kress & Specht, 2005; Specht, 2005).
MADS box candidate genes
A group of transcription factors, many of which belong to the MADS box gene family, have been shown to be involved in controlling floral organ identity in the model plants Antirrhinum majus and Arabidopsis thaliana (reviewed in Krizek & Fletcher, 2005; Theissen & Melzer, 2007). The ABC model for floral development proposes that three classes of MADS box homeotic genes are required for floral patterning in A. thaliana: A-class gene expression specifies sepal identity, A and B genes expressed together specify petal identity, B and C genes expressed together confer stamen identity, and C gene expression specifies carpel identity (Coen & Meyerowitz 1991). In A. thaliana, the A-class genes are APETALA1 (AP1) and APETALA2 (AP2) (Mandel et al., 1992; Jofuku et al., 1994), the B-class genes are represented by APETALA3 (AP3) and PISTILLATA (PI) (Jack et al., 1992; Goto & Meyerowitz, 1994), and the C-class gene is AGAMOUS (AG) (Yanofsky et al., 1990). The names of the A. majus orthologs of PI and AP3, GLOBOSA (Troebner et al., 1992) and DEFICIENS (DEF) (Schwarz-Sommer et al., 1990), respectively, have precedence; we heretofore apply these names to any Zingiberales orthologs.
It has been hypothesized that the ABC model’s definition of B-class activity in the development of petals is applicable only to the core eudicots because of the unclear homology of petals across angiosperms (Kramer & Irish, 1999). However, investigations in noncore eudicots have begun to support an expanded role for the B-class genes in the development of petals in many angiosperm lineages, despite separate evolutionary derivations of petals (Endress & Doyle, 2009). In basal eudicots Aquilegia vulgaris and Papaver somniferum (Ranunculales), B-class genes appear to be necessary for the development of both second whorl petals and third whorl stamens (Drea et al., 2007; Kramer et al., 2007). The GLO homologs from Agapanthus praecox and Elaeis guineensis, monocot flowers with petaloid inner perianth organs, have been shown to rescue the pi-1 mutant of A. thaliana (Nakamura et al., 2005; Adam et al., 2007). Zmm16, a maize (Zea mays) GLO ortholog, is also able to rescue petal development in an A. thaliana pi mutant (Whipple et al., 2004). Although data from heterologous expression studies are difficult to interpret, these results support the hypothesis that B-class genes are playing similar roles in both monocots and eudicots.
In addition, B-class mutants of maize and rice (Oryza sativa) support the homology of petals and lodicules, the second-whorl organs of grasses. Lodicules of B-class mutants are transformed into palea-like (i.e. first whorl) organs and stamens are transformed into carpel-like (i.e. fourth whorl) organs (Ambrose et al., 2000; Prasad & Vijayraghavan, 2003; Yadav et al., 2007; Yao et al., 2008). A number of morphological and gene expression studies on grasses and their closest nongrass relatives have further demonstrated the homology of lodicules and petals (Whipple et al., 2004, 2007; Sajo et al., 2008; Preston et al., 2009). Presumably it is the downstream targets of B-class genes that have been modified throughout the course of grass evolution to produce the unique morphology of lodicules (Whipple et al., 2007). It seems reasonable that the B-class genes are playing similar roles in controlling perianth and androecium identity in the Zingiberales. We hypothesize that the evolution and functional diversification of these transcription factors have contributed to the diversification of floral morphology in the Zingiberales, particularly in perianth and androecial whorls.
Changes in the Zingiberales androecium have been investigated in an evolutionary context (Rudall & Bateman, 2004), but there has been less focus on the evolution of perianth morphology. Across monocots there are examples of flowers with almost indistinguishable (i.e. undifferentiated) perianth parts (tepals) or distinguishable (i.e. differentiated) perianth parts (sepals and petals). The characters most often used to define sepals and petals are based on eudicots (Warner et al., 2008), and even so there is no single set of characters, taken in isolation, that would make an organ inherently a ‘sepal’ or a ‘petal’ (Endress, 1994). In the monocots, a fully differentiated perianth has been derived multiple times, in each case independent of eudicot sepals and petals (Endress & Doyle, 2009). In Zingiberales, the perianth is considered to be differentiated in the ginger families (Givnish et al., 1999; Ronse De Craene et al., 2003; Heywood et al., 2007), but the banana families are sometimes described as having an undifferentiated perianth (Ronse De Craene et al., 2003; Heywood et al., 2007) and sometimes as having a differentiated perianth (Givnish et al., 1999).
In this study, we examined perianth morphology in light of the Zingiberales phylogeny, investigating the evolution of character states associated with perianth differentiation. Given the importance of homeosis in the evolution of Zingiberalean flowers, particularly the development of petaloidy in the androecial whorls (Kirchoff, 1991, 1997, 1998; Kirchoff et al., 2009), we further focused on the evolution and function of the B-class gene lineages in Zingiberales. We present data on changes in copy number across the order and on the expression of GLO orthologs in two Zingiberales taxa with divergent floral morphologies. We discuss uncovered relationships between gene duplications, gene expression patterns, and changes in floral morphology.
Materials and Methods
Perianth definition and ancestral character state reconstruction
To maintain a continuity of terminology from previous literature on floral morphology in the Zingiberales, we refer to outer whorl tepals as sepals, and to inner whorl tepals as petals in all members of the order. To assess evolutionary changes in Zingiberales perianth morphology, we scored 10 characters that may differ between sepals and petals and thus may contribute to a ‘differentiated’ perianth. Outgroup taxa from all commelinid monocot orders were selected based on the most recent published phylogenies (Table 1). Recognizing that ancestral character state reconstruction is highly dependent on taxon choice (Ronse De Craene, 2008), we preferentially chose early-diverging taxa and taxa with less-derived floral morphology. Original species descriptions and accompanying illustrations, as well as descriptions, illustrations and photographs from regional flora, were used to determine character states. Character states were mapped onto a diagram of currently accepted commelinid monocot (Givnish et al., 2006; Graham et al., 2006) and Zingiberales (Kress, 1990a,b, 1995; Kress et al., 2001) phylogenetic relationships. Ancestral character states were assessed under a reversible parsimony model, as implemented in Mesquite v2.72 (Maddison & Maddison, 2009).
Table 1. Taxa scored for perianth character state reconstructions
We defined two hypothetical perianth states: one in which the perianth is completely undifferentiated and there are no discernable differences between sepals and petals, and a second in which sepals and petals differ in every character assessed. The characters we assessed were as follows. (1) size: 0, all tepals equal in size; 1, sepal and petal sizes differ (when described as ‘subequal’, sepals and petals were scored as equal in size). (2) Color: 0, sepals and petals are the same color; 1, they are different colors. (3) Texture: 0, sepals and petals have the same texture; 1, they have different textures. (4) Pubescence: 0, sepals and petals show the same pubescence patterns; 1, pubescence patterns differ. (5) Adnation: 0, sepals and petals adnate; 1, no adnation between whorls. (6) Corolla connation: 0, petals free; 1, petals connate for some to most of their length. (7) Calyx connation: 0, sepals free; 1, sepals connate. (8) Shape: 0, sepals and petals of approximately the same shape; 1, their shapes differ. Zygomorphy within calyx (9) and corolla (10): 0, whorl actinomorphic; 1, whorl zygomorphic. We also coded staminode characteristics (11): 0, presence; 1, absence; 2, petaloid staminodes; 3, staminodial labellum.
Perianth character states (characters 1 to 10) for each taxon were summed to yield a ‘dimorphism score’. A completely undifferentiated perianth (dimorphism score of 0) would have identical sepals and petals, fused (adnate) into a single floral tube. A fully differentiated perianth (dimorphism score of 10) would have sepals and petals that differed in size, color, shape, pubescence and texture and were fused into separate calyx and corolla tubes. Calyx and corolla zygomorphy (characters 9 and 10) were included to account for perianth morphology in some Zingiberales and Commelinales. In Musaceae and many Pontederiaceae a single petal is differentiated; in Heliconiaceae, a single sepal is differentiated. The single tepal most commonly differs from the other tepals in either size or color, or both (see references included in Table 1); however, these organs were not considered when scoring size, shape and color differences. Rather, their differentiation was included in the zygomorphy characters. The dimorphism score was reconstructed as both a discrete and a continuous character and the results compared.
RNA extraction and cDNA synthesis
Taxa were selected to represent floral diversity within each family of the order Zingiberales (Table 2). Floral material was preserved in RNAlater (Ambion, Austin, TX, USA). After one night of incubation at 4°C and up to 2 wk at −20°C, tissue in RNAlater was archived at −80°C. RNA was extracted from floral material representing a range of developmental stages using Plant RNA Extraction Reagent (Invitrogen, Carlsbad, CA, USA), following the manufacturer’s protocol. The RNA pellet was re-eluted in 10–30 μl of water and also stored at −80°C. RNA was quantified using a NanoDrop 1000 spectrophotometer (Thermo Scientific, Waltham, MA, USA). DNA was removed from 2 μg of RNA with RQ1 RNAse-free DNase (Promega, Madison, WI, USA) and reverse transcribed into cDNA using a polyT primer and Moloney Murine Leukemia Virus (M-MLV) reverse transcriptase (Promega, Madison, WI, USA) following the manufacturer’s protocols. The success of the reverse transcription reaction was assessed by amplifying β-actin from the cDNA, using the following intron-spanning primer pair: ACT-F, GGA CGA ACA ACT GGT ATC GTG CTG, and ACT-R, GAT GGA TCC TCC AAT CCA GAC ACT GTA.
Table 2. Taxon sampling for phylogenetic analysis of GLO homologs in the Zingiberales
aLocation of live accessions or herbarium sheets: Lyon Arboretum, Oahu, Hawaii, USA; McBryde Botanical Garden, Kauai, Hawaii, USA; University of California Botanical Garden (UCBG); University of California Berkeley Herbarium (UC), Smithsonian Greenhouses (NMNH).
cDNA was diluted 1 : 10 in water, and 2 μl of the dilution used in 20-μl PCR reactions containing 0.5 pmol each of the forward and reverse primers, 0.4 U iProof DNA polymerase (Bio-Rad, Hercules, CA, USA), 4 μmol dNTPs and 1 μg of bovine serum albumin (BSA). The final MgCl2 concentration was 2.5 mM. We achieved limited success in amplifying GLO-like genes using forward primers degenerate for the MADS domain and a polyT reverse. Consequently, we designed less degenerate primers after comparing GLO-like sequences across commelinid monocots. Multiple primer pair combinations were used for each cDNA sample: ZinGLO-F, CAG GTS ACC TTC TCC AAG C and ZinGLO-R, AGG TTD GGY TGG YTG GGT TG; ssMADS-F, CAR GTK ACC TTC TGC AAG and 3′-GLO, CAT ATA AGT CAG TTG CTT GTT CTC CTC CTC; ZinGLO-R2, GGY TGS TSR CGG AAG GCC AT was used in combination with ZinGLO-F and ssMADS-F. The full volumes of the PCR reactions were run out on 1.2% agarose gels and bands of the appropriate size were excised and cloned. Between four and 20 clones were picked from each cloning reaction and sequenced using vector-specific primers. The number of colonies sequenced for each family ranged from 25 (Lowiaceae) to 140 (Costaceae). In order to saturate gene recovery for at least two species across the order, 100 Costus spicatus colonies and 98 Musa basjoo colonies were sequenced. In addition, ZinGLO clade-specific primers were designed and used on taxa where particular paralogs had not been recovered. The primer combinations used in this serial PCR approach were: ZinGLO1-F, GAG TAC TGC AGC CCA TCC AC; ZinGLO1-R CAA TTC CTT AGG GTT GAG AGA A; ZinGLO2-F, ATC AAG AAG GCG AGG GAG AT; ZinGLO2-R, AAG TTC CTT GGG ATA AAG CGA G; ZinGLO3-F, GGG ATC ATT AAG AAA GCG AGA GAA; ZinGLO4-F, CAT ATT CTC AAG CTC TGG C and ZinGLO3&4-R, TGG AAT TAA TTC TTT TGG G. All sequences were obtained using BigDye v3.1 on an ABI 3100 Genetic Analyzer (Applied Biosystems, Foster City, CA, USA). All sequences were deposited in GenBank (accession numbers GU594899–GU594995).
Multiple sequence alignment and phylogenetic analysis
Zingiberales GLO-like sequences have been shown to form a clade (posterior probability (pp) = 0.91) within a broader monocot GLO-like phylogeny (Mondragon-Palomino et al., 2009). For this reason, we included only commelinid monocot GLO homologs (retrieved from GenBank) as outgroups in our analysis. Nucleotide sequences were translated into protein using MacClade v4.08 (Maddison & Maddison, 1998) and aligned using Muscle (Edgar, 2004). The resulting protein alignment was transferred manually to the nucleotide data and the final alignment edited by hand. Once primer sequences had been removed, the final alignment was 597 nucleotides in length.
Model selection was performed using MrModeltest (Nylander, 2004). The alignment was partitioned into first and second vs third codon positions and each partition assessed. No difference in model selection was identified for the two partitions (GTR + I + G selected for all partitions and the unpartitioned data set). When the data were partitioned according to protein domain (M, I, K and C), MrModeltest selected separate models for each partition (Table 3). The resulting partitioned data set was analyzed using Bayesian inference of phylogeny, as implemented in MrBayes v3.2.1 (Huelsenbeck & Ronquist, 2001). Gaps in the alignment were treated as missing data. Two analyses were run in parallel until both converged on similar log likelihood scores (average standard deviation of split frequencies < 0.01). The log likelihood scores reached a plateau after c. 10% of the generations completed (assessed using Tracer; Rambaut & Drummond, 2007). Consequently, the first 10% of the trees were discarded as ‘burnin’ and a 50% majority rule tree was constructed from the remaining trees. Maximum likelihood searches using the unpartitioned data set and 500 ML bootstrap replicates were performed using Garli (Zwickl, 2006) on the CIPRES web portal (Miller et al., 2009). A consensus of bootstrap trees was constructed using the sumtrees script in Dendropy (Sukumaran & Holder, 2009). Trees were edited using Mesquite (Maddison & Maddison, 2009), Dendroscope (Huson et al., 2007) and Illustrator CS4 (Adobe).
Table 3. Models of nucleotide evolution selected for GLO-like data set partitioned according to protein domain
GTR + G
SYM + I + G
GTR + I + G
HKY + I + G
RNA in situ hybridization
Expression of ZinGLO1 was assessed in Musa basjoo Siebold (Musaceae) and Costus spicatus Swartz (Costaceae). The expression of ZinGLO2 was assessed in C. spicatus. In situ hybridizations were performed as described in Bartlett et al. (2008). The RNA probes for CsGLO1 and CsGLO2 included the MADS box, the I region, the K region and a section of the C region (MIKC) of the MADS box genes (GenBank GU594899 and GU594931). We are confident that there was no cross-hybridization between probes because of distinct expression patterns obtained using similarly designed probes for CsGLO1, CsGLO2 and a C. spicatus AGAMOUS ortholog (data not shown). The Musa probe (GenBank GU594929) was designed to exclude the MADS box.
Floral organs were dissected from C. spicatus flowers shortly before anthesis. RNA extraction and cDNA synthesis were performed as described above. Primers were designed to flank multiple introns and to amplify different sized bands for the C. spicatus ZinGLO1 ortholog (CsGLO1-F, GAG TAC TGC AGC CCA TCC AC; CsGLO1-R, CAA TTC CTT AGG GTT GAG AGA A) and the ZinGLO2 ortholog (CsGLO2-F, ATC AAG AAG GCG AGG GAG AT; CsGLO2-R, AAG TTC CTT GGG ATA AAG CGA G). Reagent concentrations were as described in ‘Amplification of GLO homologs’. Cycling conditions were as follows: initial denaturation at 98°C for 4 min, 28 amplification cycles (98°C for 20 s, 62°C for 20 s and 72°C for 30 s), and a final extension step of 72°C for 7 min. β-actin was amplified from all tissues as a reference. All bands were sequenced to verify identity. One PCR product from each of the CsGLO1 and CsGLO2 RT-PCR experiments was cloned and multiple clones sequenced to verify that the bands represented single sequences.
Character evolution in the Zingiberales and the broader commelinid monocots
Perianth morphology in the Zingiberales is diverse (Fig. 1). In the banana families, the perianth provides the majority of the floral display (Fig. 1b–d,f,g,j). In the ginger families, it is the petaloid staminodes that are large and brightly colored, while the perianth is often inconspicuous (Fig. 1e,h,i,k–p). Ancestral character state reconstruction of major features of perianth morphology helps in the assessment of the characters contributing to perianth differentiation (Fig. 2).
Perianth dimorphism Both hypothetical morphologies (a dimorphism score of 0 or 9) are rare in the commelinid monocots, although both do occur (Fig. 3). Flowers of Musa have a perianth dimorphism score of 1. The only distinction between the perianth whorls lies in the single free, larger adaxial petal, resulting in zygomorphy of the petal whorl (character 10; Fig. 2j). Although a completely differentiated perianth is not ancestral in the Zingiberales, this reduced perianth dimorphism appears to be a derived condition in Musaceae (Fig. 3). The adnation between the sepal and petal whorls observed in Musaceae and Heliconiaceae is derived separately in each family (Fig. 2f). The perianth dimorphism scores of Lowiaceae, Strelitziaceae, Zingiberaceae, Cannaceae and Marantaceae are all comparable, lying between 5 and 6. This similarity in scores, however, does not necessarily imply homology of the differentiated perianth in these families, as similar scores result from the presence or absence of different characters.
The reconstruction of dimorphism score as a continuous character summarized the results of individual reconstructions fairly adequately, while reconstructing dimorphism as a discrete character obscured much of the complexity revealed by the individual character state reconstructions (data not shown). As such, the results and discussion will be restricted to the continuous character reconstruction. The rounded scores at most nodes correspond to the results obtained by assessing each character individually (Figs 2, 3). Using the continuous reconstruction, we can trace the gradual accretion and subsequent losses of morphological differences between sepals and petals in the commelinids.
The perianth dimorphism score both at the base of Commelinales plus Zingiberales and at the base of the commelinid monocots is 3. This score is in agreement with the reconstruction of individual character states (Fig. 3). The ancestral commelinid monocot perianth is reconstructed as being differentiated in shape, not a single floral tube, and may or may not have been differentiated in size. This score was consistent regardless of the phylogenetic hypothesis used to reconstruct character history in the commelinid monocots (Chase et al., 2006; Givnish et al., 2006; Graham et al., 2006). The perianth dimorphism score at the base of the Zingiberales is also 3, in accordance with the results from reconstructing individual characters (Figs 2, 3). A zygomorphic corolla, shape differences between outer and inner whorl tepals and an absence of adnation are the inferred ancestral conditions for the Zingiberales. Size difference between sepals and petals is reconstructed as equivocal.
The well-differentiated perianth, a characteristic of flowers in the ginger families, Commelinaceae, Arecales and the basal Poales (i.e. dimorphism scores > 5; Fig. 3), appears to have been independently derived in these four groups. Zygomorphy in the corolla was reconstructed as ancestral in the Zingiberales (Fig. 2j), as has been found for the androecium in a separate investigation (Rudall & Bateman, 2004).
Stamen whorls The presence of staminodes is not reconstructed as ancestral in the Zingiberales (Fig. 2k). Petaloid staminodes probably evolved on the branch leading to Heliconiaceae plus the ginger families (Fig. 2k).
‘Labellum’ is a term used to describe nonhomologous floral organs in divergent taxa (Rudall & Bateman, 2002). Many of the families of the Zingiberales have been described as possessing a labellum, all of varying homologies (Eichler, 1878; Thompson, 1933; Kirchoff, 1983, 1988b, 1997, 1998; Kirchoff & Kunze, 1995). The labellum of Costaceae and Zingiberaceae, however, arises from the fusion of multiple petaloid staminodes (Fig. 1). It is this novel compound organ to which we refer when we use the term ‘staminodial labellum’. The staminodial labellum is reconstructed to have evolved before the divergence of Costaceae and Zingiberaceae (Fig. 2k)
Gene tree: homology, duplications and losses
Phylogenetic reconstruction of GLO-like gene evolution in the Zingiberales revealed a complex history of gene duplications and losses. We identified at least four GLO homologs in the Zingiberales (Fig. 4). These may be the result of Zingiberales-specific duplications: all GLO homologs form a clade to the exclusion of other monocot sequences, but the Zingiberales GLO-like clade received only moderate support (pp = 0.81). Support levels are low throughout the Zingiberales, particularly at deeper nodes.
Bayesian phylogenetic analysis resolved three clades and one grade of GLO homologs (Fig. 4). Two of the three ZinGLO lineages, ZinGLO3 and ZinGLO4, form distinct clades (pp = 0.88 for both clades). The ZinGLO1 clade is only weakly supported (pp = 0.56), although internal branching patterns are consistent with broader organismal phylogeny. We have designated the grade leading to the ZinGLO3/ZinGLO4 node as ZinGLO2. Each of the well-supported clades (pp = 0.93–1.00) within the ZinGLO2 grade show an internal branching structure consistent with the Zingiberales organismal phylogeny. Because there is no evidence for further duplications within any of the ZinGLO2 clades, the grade could more accurately represent a single GLO homolog. We tested the hypothesis that the ZinGLO2 grade could form a monophyletic GLO-like gene lineage by repeating the phylogenetic analysis, constraining the ZinGLO2 grade to be monophyletic. The evidence against the constrained model was assessed using the Bayes factor (Kass & Raftery, 1995; Nylander et al., 2004). The harmonic mean of log likelihoods in the constrained analysis (M0), as estimated by MrBayes, equaled −21 210.14; that in the unconstrained analysis (M1) equaled −22 044.54. The Bayes factor10 thus equaled 1.04. A Bayes factor between 1 and 3 implies very little evidence against M0 (Kass & Raftery, 1995; Nylander et al., 2004). We therefore did not reject the hypothesis that the ZinGLO2 grade is monophyletic. In all further discussion, we refer to the genes belonging to this putative clade as orthologs of ZinGLO2.
Nucleotide sequence divergence between individual ZinGLO paralogs is low (81.2–95.9%), but the evidence supports our claim that the clades represent paralogs rather than alleles. Both alleles and paralogs can be discerned in the ZinGLO phylogeny. For example, we uncovered multiple alleles of both ZinGLO2 and ZinGLO1 from Strelitzia reginae (Strelitziaceae), Halopegia azurea (Marantaceae), and C. spicatus (Costaceae). Similarly, multiple alleles of ZinGLO3 and ZinGLO4 were recovered from Globba laeta (Zingiberaceae) (Fig. 4). Sequence divergence is particularly low between ZinGLO3 and ZinGLO4 (90–95.9% similarity). To confirm that these are indeed paralogs rather than persistent alleles, we sequenced intron 6 of these genes. Intron 6 of ZinGLO3 is consistently 80 bp shorter than intron 6 of ZinGLO4 (data not shown). This adds support to the hypothesis that the ZinGLO3 and ZinGLO4 clades represent separate GLO paralogs.
To investigate the order of inferred gene duplications, we scored each of the families in the Zingiberales for the number of GLO-like genes retrieved (Fig. 5). Commelinaceae, the only family in the Commelinales from which GLO homologs have been sequenced, was used as the outgroup. Two GLO-like genes are recorded from Tradescantia reflexa, but only a single copy has been retrieved from Commelina communis (Ochiai et al., 2004). Commelinaceae was therefore coded as polymorphic (one or two GLO-like genes). When gene copy number was reconstructed as a discrete character using parsimony (Maddison & Maddison, 2009), the common ancestor of Zingiberales was found to have a single GLO homolog (Fig. 5a). The ancestor at the node separating Musaceae from the remaining Zingiberales was reconstructed to have three GLO homologs. Tracing the history of individual ZinGLO homologs yielded similar results (Fig. 5b–e). ZinGLO2 was gained after the divergence of Musaceae from the remainder of the Zingiberales. ZinGLO3 was either gained at the same point and subsequently lost in the Strelitziaceae, or, equally parsimoniously, was gained independently in Lowiaceae and before the divergence of Heliconiaceae from the ginger families. This second hypothesis of two separate gains seems unlikely considering the low sequence divergence between ZinGLO3 orthologs (92.2–99.8% similarity; Fig. 5d): the Lowiaceae ZinGLO3 sequences are nested within a clade of ginger ZinGLO3 sequences, indicating common origin. These results imply that two separate gene duplication events occurred along the branch directly after the divergence of Musaceae from the remaining Zingiberales.
ZinGLO4 appears to be the result of a gene duplication event before the divergence of Costaceae and Zingiberaceae. The presence of ZinGLO4 in only these two families, coupled with the extreme similarity of ZinGLO4 orthologs from Costaceae and Zingiberaceae (96.3–99.5% similarity), adds credence to this reconstruction (Fig. 5e).
ZinGLO1 expression in Musa basjoo and Costus spicatus
Plants in the Musaceae are monoecious, with separate female, transition, and male flowers produced on the same inflorescence but at different times during development. Transitional flowers are functionally male, but have gynoecium morphology intermediate between male and female flowers (Simmonds, 1966). Female flowers in M. basjoo possess five staminodes in place of fertile stamens, and male flowers have a reduced, aborted gynoecium. Early floral development appears to be similar in male and female flowers, and differentiation between the female, transition and male flowers occurs later in development (White, 1928; Fig. 1b,c). The flowers examined in this study were all male. White (1928) observed that the floral meristems start as rounded domes and become flattened during the course of development. Sepals are initiated first, and the petals are formed in the gaps between sepals. The three outer whorl stamens are antisepalous and form internally to the perianth whorls. Inner whorl stamens are antipetalous and appear to initiate after the outer whorl stamens. The staminode in the accession of M. basjoo we examined appears to arise from the anterior side of the adaxial free petal (Fig. 6a). The gynoecium is inferior and initiates last.
Orthologs of ZinGLO1 were not detected in early floral meristems of M. basjoo or C. spicatus (Fig. 6g–h). In M. basjoo, MbGLO1 RNA was detected at high levels in the region internal to the developing sepals of young flowers (Fig. 6b). It is unclear whether petals and stamens in M. basjoo originate from common petal-androecium primordia, which occur frequently in the Zingiberales (Kirchoff, 1983, 1988b, 1997; Kirchoff & Kunze, 1995; Kirchoff et al., 2009), or from individual organ primordia. As such, the observed MbGLO1 expression may have been in common petal-androecium primordia, or in individual androecial and/or petal primordia. Petals in M. basjoo are always formed opposite stamens, not opposite other petals (Fig. 6a), and identical expression patterns were observed in multiple sections at varying levels. Therefore, whether or not stamens and petals arise from common primordia, MbGLO1 expression was detected in developing petals and stamens. Expression in older flowers was detected in the five fertile stamens, the adaxial staminode and the free petal (Fig. 6d).
For C. spicatus, the development of a closely related species, Costus scaber, has been well characterized (Kirchoff, 1988b) and was useful in determining when and where ZinGLO orthologs were expressed. The perianth of C. spicatus consists of three connate sepals and three petals fused proximally into a floral tube, but free distally. Five infertile staminodes are fused to form the abaxial labellum, with a single adaxial fertile stamen (Fig. 6d). Early in development, expression of CsGLO1 was detected in the common petal-androecium primordia (Fig. 6e). Expression later became restricted to the fertile stamen, the labellum and the petal margins (Fig. 6f). Based on results from semiquantitative RT-PCR in flowers just before anthesis, CsGLO1 expression was strongest in the labellum, but was also found in petals, stamens, and the gynoecium (Fig. 6p).
ZinGLO2 expression in Costus spicatus
The ortholog of ZinGLO2 was not retrieved from Musa. Expression of CsGLO2, the C. spicatus ortholog of ZinGLO2, was not detectable in early floral meristems (Fig. 6n). Expression became evident once sepals and the common petal-androecium primordia had differentiated (Fig. 6k). At this stage, CsGLO2 was expressed throughout the developing flower at a higher level than that observed for CsGLO1, a result corroborated with RT-PCR (Fig. 6p). CsGLO2 expression was strongest in the common androecium-petal primordia and weakest in the sepals. Expression in the sepals was undetectable using in situ hybridization in mature flowers (Fig. 6l–m). RT-PCR indicated that CsGLO2 was expressed weakly in the sepals, with stronger expression in the petals, stamens, labellum and gynoecium of mature flowers.
Perianth evolution in the commelinid monocots
The ancestral perianth character state for monocotyledons has been reconstructed as ‘undifferentiated’ in several analyses (Ronse De Craene et al., 2003; Zanis et al., 2003; Endress & Doyle, 2009). A differentiated perianth is likely to have evolved multiple times in the monocots (Ronse De Craene et al., 2003; Zanis et al., 2003; Endress & Doyle, 2009). In Ronse De Craene et al.’s (2003) analysis, the character state at the base of the Zingiberales was reconstructed as equivocal, but Strelitzia, Orchidantha and Heliconia were all scored as undifferentiated and phylogenetic relationships among the Zingiberales families were not well resolved (Ronse De Craene et al., 2003). Givnish et al. (1999) reconstructed the ancestral perianth state in commelinids as differentiated. Their perianth character states for Musaceae and Phylidraceae were in conflict with those of the current analysis, but even if the character states and the tree topology were adjusted to be congruent with our analysis, the ancestral state of the commelinid monocots and the Zingiberales would still be reconstructed as ancestrally differentiated. The terms ‘differentiated’ and ‘undifferentiated’ are, however, imprecise and ultimately obscure the exact changes in morphology that have occurred through the course of evolution.
In our analysis, a fully differentiated perianth was not reconstructed as ancestral in the commelinid monocots. At least three separate derivations of a well-differentiated perianth were reconstructed: one in the Zingiberales, one in the Commelinales, and either a single derivation at the base of the Arecales plus Poales, or a separate derivation in each order if they are not sister (Chase et al., 2006; Graham et al., 2006). The results of our individual character state and dimorphism score reconstructions suggest that the well-differentiated perianth characteristic of Lowiaceae, Strelitziaceae and the ginger familes, especially Costaceae, was not derived in a single saltatory event, but rather may have been derived through the gradual accretion of differences between outer and inner whorl tepals (Figs 2, 3, 7). This could be interpreted as the progressive partitioning and canalization of the perianth into outer and inner whorl organs (Flatt, 2005).
ZinGLO gene duplications and losses
Our results indicate three duplication events giving rise to four GLO homologs in the Zingiberales. There are low levels of nucleotide divergence between all four GLO homologs, particularly between ZinGLO3 and ZinGLO4. The level of divergence between ZinGLO3 and ZinGLO4 within a single species is similar to that observed between pairs of duplicate maize MADS box genes generated in an allotetraploidy event that occurred between 11 and 21 Mya (Mena et al., 1995; Theissen et al., 1995; Gaut & Doebley, 1997; Cacharron et al., 1999; Munster et al., 2001). Ancestral character state reconstructions put the ZinGLO3-ZinGLO4 duplication on the branch leading to Zingi-beraceae plus Costaceae (Figs 5, 7). These families are estimated to have diverged from each other c. 105 Mya (Kress & Specht, 2006).
ZinGLO4 and ZinGLO3 may be confoundingly similar for a number of reasons: they may have arisen in a more recent duplication event than our analysis suggests; these GLO paralogs may be under strong purifying selection; or the observed similarity may be attributable to gene conversion. There is evidence for extensive gene conversion in rice (Wang et al., 2007). The assessment of nucleotide sequence divergence highlights the importance of phylogenetic analysis and dense taxonomic sampling in determining gene homologies, especially in the absence of a sequenced genome. Because of the high degree of similarity between ZinGLO paralog sequence divergence (81.2–95.6%) and within-ortholog sequence divergence (84.9–99.8%), a simple BLAST search would not have been able to discern orthologs, paralogs and alleles.
There is evidence for multiple losses of GLO-like genes in the order. These ‘gene losses’ may have been sampling artifacts rather than true losses, but this seems unlikely as the same sampling strategy was used on all taxa investigated. The primer combinations we used also amplified DEF-like and AP1-like genes, indicating broad amplification of closely related MADS box gene lineages. Although nucleotide divergence between individual ZinGLO paralogs is low, the possibility remains that these ‘lost’ paralogs have diverged so extensively in their nucleotide sequences that our primers were unable to amplify them. Alternatively, they may be expressed at extremely low levels as compared with their orthologs in different taxa and thus remained undetected. Assuming a common origin of ZinGLO3, losses occur in three places on the tree: ZinGLO1 was lost in Zingiberaceae (Fig. 5b), and ZinGLO3 was lost in Strelitziaceae and Cannaceae. ZinGLO3 also may have been lost in specific lineages within Costaceae. Although ZinGLO3 was repeatedly retrieved from Tapeinochilos and Dimerocostus, attempts to amplify this gene from Costus were unsuccessful.
ZinGLO gene duplications are associated with increasing perianth dimorphism
A central concept in evo-devo is the hypothesized role of the diversification of transcriptional regulation (Levine & Tjian, 2003; Wray et al., 2003). The raw material for transcriptional diversification is thought to be duplicate genes. Genes may be duplicated by a whole-genome duplication event (polyploidy), a tandem duplication or transposition-duplication (reviewed in Freeling, 2009). These maintained duplicates may decay and become pseudogenes, they may retain their ancestral functions and expression patterns or they may subfunctionalize or neofunctionalize over time, leading to phenotypic novelty (Ohno, 1970; Lynch et al., 2001; Freeling & Thomas, 2006; Freeling, 2009).
GLO-like gene duplication and diversification may have contributed to floral diversification in the Zingiberales. ZinGLO gene duplications are reconstructed to have occurred on the branch following the divergence of Musaceae and before the divergence of Costaceae and Zingiberaceae. Perianth dimorphism score remains constant at a level of 3 from the origin of the commelinid monocots to the base of the Zingiberales, but begins to increase following the divergence of Musaceae from the ancestor of the remaining Zingiberales. The first two hypothesized gene duplication events coincide with the beginning of this gradual increase in perianth dimorphism (Fig. 7). Gene duplication and subsequent subfunctionalization may have allowed for the partitioning and subsequent differentiation of the perianth. It has been hypothesized that, rather than directly controlling petal identity, B-class genes define a particular region in the developing flower (Irish, 2009). Increasing B-class gene diversity may allow for an increased number of floral regions, resulting in increased modularization and diversification of the perianth (Mondragon-Palomino & Theissen, 2009). CsGLO1 is expressed early in common petal-stamen primordia, and later only in the androecium and gynoecium. CsGLO2 is expressed early in the sepals and common petal-stamen primordia, and later in the petals, androecium and gynoecium. These overlapping but not identical expression domains may be the result of subfunctionalization or neofunctionalization of the duplicated ZinGLO genes, allowing for the formation of new floral regions and ultimately further differentiation of sepals and petals.
If the orchid labellum is interpreted as an elaborated inner whorl tepal (see Rudall & Bateman, 2002 for a discussion), perianth morphologies in the Heliconiaceae, Musaceae, Pontederiaceae, and Orchidaceae share some striking similarities. A perianth consisting of five tepals, with a single differentiated posterior tepal, occurs in all of these families. It has been hypothesized that this within-whorl modularization in Orchidaceae may have been a result of duplications and subsequent subfunctionalization and neofunctionalization in the DEF gene lineage (Mondragon-Palomino & Theissen, 2008, 2009; Mondragon-Palomino et al., 2009). Intriguingly, MbGLO1 was found to be expressed in the single free petal of M. basjoo, while it was not detected in the remaining perianth members (Fig. 6c). Modularization within individual perianth whorls, mediated by B-class MADS box gene evolution, may be a trend in these more derived monocot lineages.
ZinGLO gene duplications and the evolution of the androecium in the Zingiberales
The staminodial labella in Costaceae and Zingiberaceae are probably homologous organs. There have been no studies of floral development in the earliest diverging Zingiberaceae lineages, the monotypic subfamilies Tamijieae and Siphonochiloideae (Kress et al., 2002), but in both subfamilies the well-developed lateral staminodes are fused to the labellum, as is the case in Costaceae (Kirchoff, 1988b; Sakai & Nagamasu, 2000; Kress et al., 2002). It remains unclear whether the labella in Tamijia and Siphonochilus are the product of the fusion of four, or five, staminodes. The labella are, however, bilobed in both genera (Sakai & Nagamasu, 2000; Kress et al., 2002), as in other Zingiberaceae where the anterior staminode initiates in a position confluent with the rest of the labellum but ultimately aborts (Kirchoff, 1997, 1998; Box & Rudall, 2006).
A single gene duplication event was reconstructed to have occurred in the common ancestor of Costaceae plus Zingiberaceae, leading to the presence of ZinGLO4 exclusively in these two families. This hypothesized gene duplication occurred concurrently with the reconstructed derivation of the staminodial labellum. By contrast, the acquisition of petaloid staminodes is not associated with any detected GLO-like gene duplication event. This does not preclude the involvement of GLO-like genes in the evolution of petaloid staminodes: the possibility remains that ZinGLO2 and ZinGLO3 have been modified through the course of evolution in the lineage that led to Heliconiaceae plus the ginger families, but not in Strelitziaceae or Lowiaceae. Expression of a ZinGLO3 ortholog has been assessed in Alpinia oblongifolia (Zingiberaceae) (Gao et al., 2006). As there is a fairly low degree of sequence divergence between ZinGLO3 from A. oblongifolia and ZinGLO4 from A. hainanensis (90.8% similarity), in situ hybridization studies in A. oblongifolia may well have been detecting the expression of both ZinGLO3 and ZinGLO4. Expression was detected early on in the common petal-androecium primordia, later in the petals and the androecium. This expression pattern is very similar to that observed for CsGLO2, except CsGLO2 is expressed in the sepals early in development. The expression of these homologs needs to be investigated more fully across the order.
If the acquisition of ZinGLO4 was one of the key events associated with the evolution of the staminodial labellum, it may be the case that ZinGLO4 is the ‘labellum gene’ and is expressed and functions only in the labellum, or it may show a wider expression pattern. It may be the combination of multiple GLO homologs, rather than expression of a single ortholog, that confers labellum identity (Fig. 7). Considering the combinatorial functioning of MADS box genes (Egea-Cortines et al., 1999; Honma & Goto, 2001; Theissen, 2001), and the broad expression patterns observed for ZinGLO1, ZinGLO2, ZinGLO3 and possibly ZinGLO4 (Gao et al., 2006), the combination hypothesis seems more likely. There may be stoichiometric competition between GLO paralogs for binding partners in higher-order MADS complexes. A certain combination of MADS tetramers in a certain ratio is what is necessary for conferring organ identity (Fig. 7b). There is evidence for the combination hypothesis in Petunia hybrida, where dosage effects have been observed for B-class genes (Vandenbussche et al., 2004). There is suggestive evidence for differences in expression levels between paralogs in C. spicatus (this study) and Ranunculaceae (Kramer et al., 2007; Rasmussen et al., 2009). Petal-specific expression of AP3-3 is seen in A. vulgaris, but none of the AP3 or PI homologs in A. vulgaris has been found to be expressed only in the staminodia (Kramer et al., 2007). Similarly, none of the orchid DEF homologs is expressed exclusively in the tepaloid labellum, but it is perhaps the combination of all four that is necessary for labellum development (Mondragon-Palomino & Theissen, 2009). We plan to explore and test this hypothesis in the Zingiberales by examining the expression of all four ZinGLO paralogs across the order.
In conclusion, we have uncovered multiple gene duplication events within the GLO gene lineage in the Zingiberales. These events are separately associated with the increased modularization of the perianth and the acquisition of the staminodial labellum. In addition, the GLO-like gene lineage has diversified in association with the increased differentiation of the perianth. These results provide suggestive evidence that GLO gene family evolution has contributed to floral morphological evolution in the Zingiberales.
This work was supported by US National Science Foundation awards to C.D.S. (IOS 0845641) and M.E.B. (DEB 0808298), the South African National Research Foundation, the Botanical Society of America and The Heliconia Society International. The authors would like to thank Ana Almeida, Katrina Hong, Sankar Sridaran and Solomon Stonebloom for assistance with primer design, gene recovery and RNA and DNA extractions, Bruce Kirchoff for thoughtful discussion and the Specht lab for comments on the manuscript.