Evolution of unisexual flowers in grasses (Poaceae) and the putative sex-determination gene, TASSELSEED2 (TS2)


Author for correspondence: Elizabeth A. Kellogg Tel: +1 314 516 6217 Fax: +1 314 516 6233 Email: tkellogg@umsl.edu


  • • Unisexuality has evolved repeatedly in flowering plants, but its genetic control is not understood in most cases. In maize (Zea mays), unisexual flower development is regulated by a short-chain dehydrogenase/reductase protein, TASSELSEED2 (TS2), but its role in other grass lineages is unknown.
  • • TS2 was cloned and sequenced from a broad range of grasses and compared to available sequences from other flowering plants using phylogenetic analysis and tests for selection. Gene expression was investigated using reverse transcriptase–polymerase chain reaction (RT-PCR) and in situ hybridization.
  • • TS2 orthologs appear to be restricted to monocots. The TS2 protein sequence was found to be generally under purifying selection in bisexual and unisexual lineages alike. Only one site, in unisexual herbaceous bamboos, is potentially under positive selection. TS2 was expressed broadly in all sampled tissues of unisexual and bisexual grasses, and was also expressed in rice flowers in floral organs that do not abort.
  • • TS2 may have a more general developmental role in most grasses than programmed cell death of the developing gynoecium, but has been co-opted to this role within a subset of Poaceae, probably as a result of alterations in the activity or regulation of other genes in the gynoecial pathway.


Flowering plants exhibit a diverse array of sexual systems (reviewed in Barrett, 2002; Tanurdzic & Banks, 2004). Although most species produce only bisexual flowers, with functional staminate (pollen-producing, often called male) and pistillate (ovule-producing, often called female) organs in the same flower, an estimated 30% of species produce at least some unisexual flowers (Richards, 1997). These unisexual species are distributed throughout the angiosperm phylogeny, indicating that dicliny (the presence of unisexual flowers) has evolved repeatedly, likely through diverse mechanisms (Charlesworth, 1985; Mitchell & Diggle, 2005).

Evolution of sex expression in grasses

Transitions between bisexual and unisexual flowers have occurred frequently in the 70 million year history of the Poaceae. The grass outgroups in the graminoid Poales include a mixture of unisexual and bisexual taxa (Stevens, 2001 (onwards)). The earliest diverging member of the graminoids, Flagellaria, is bisexual, the next branch leads to Restionaceae and Anarthriaceae, which are both unisexual, and the next branch leads to Joinvillea, with bisexual flowers. The sister to Poaceae is now thought to be the western Australian family Ecdeiocoleaceae (Michelangeli et al., 2003), in which flowers are exclusively unisexual (Rudall et al., 2005). The earliest branch of the Poaceae includes Anomochloa and Streptochaeta, both of which are bisexual, but the next lineage leads to present-day Pharoideae, in which all species have exclusively unisexual flowers. These shifts between different forms of sex expression have continued throughout the history of the family, and estimation of the ancestral condition for the family is highly sensitive to assumptions about branch lengths and transition probabilities (data not shown). Even among closely related species, sex expression is highly variable; at least six transitions in sex expression have been estimated in the chloridoid genus Bouteloua alone (Columbus, 1999; Kinney et al., 2003). Phylogenetic data thus show that changes in sex expression occur frequently, and gain or loss of unisexuality is apparently common.

It is convenient to describe grass taxa as being unisexual or bisexual, but this oversimplifies the variation in sex expression, and indeed almost all possible breeding systems can be found in the grasses (Connor, 1981; Chapman, 1990; Watson & Dallwitz, 1992). Dioecy (staminate and pistillate flowers on separate plants) occurs in subfamilies Chloridoideae, Pooideae, and Danthonioideae; in the latter two subfamilies it is often associated with apomixis. Monoecy also occurs, with the most familiar example being maize (Zea mays). Among monoecious species, the staminate and pistillate flowers may be borne in separate inflorescences (e.g. maize), in separate spikelets in the same inflorescence (e.g. Pharoideae, tribe Olyreae in the Bambusoideae, and Tripsacum, Coix, and Heteropogon in Panicoideae), or in separate flowers in the same spikelet (e.g. Ixophorus unisetus). Other grasses are andromonoecious, with bisexual and staminate flowers in the same spikelet (e.g. most members of Panicoideae). Gynomonoecy, with bisexual and pistillate flowers in the same spikelet, occurs sporadically among the grasses (Connor, 1981); generally spikelets contain multiple flowers, the lower of which are bisexual and the upper of which often produce only a gynoecium.

Sterile flowers are also common and may co-occur with bisexual or unisexual flowers. In some cases these appear to be simply underdeveloped, perhaps as a result of environmental conditions. In other cases, sterility is clearly genetically and developmentally imposed. Examples of the latter include the lower two flowers of the spikelet in Oryza sativa. These are reduced to empty lemmas, and often called ‘glumes’ in the literature; the true glumes are reduced to tiny flaps, sometimes known as rudimentary glumes.

In all studied species of Poaceae and their relatives, flowers initiate both gynoecia and androecia (Cheng et al., 1983; Le Roux & Kellogg, 1999; Zaitchik et al., 2000; Rudall et al., 2005) even if the mature flower will be unisexual (Fig. 1). Staminate flowers form when the gynoecium fails to develop fully, and pistillate flowers when the androecium fails to mature. This pattern of selective organ abortion is also found in unisexual flowers in many other families, including Actinidiaceae, Asparagaceae, Caryophyllaceae, Curcubitaceae, Fabaceae and Polygonaceae (reviewed in Ainsworth, 2000). Control of sex expression is thus frequently the control of floral organ development, stopping or starting an existing, functional developmental pathway.

Figure 1.

Scanning electron micrographs of developing flowers in selected grasses. (a) A bisexual flower in Coelorachis aurita. Reproduced from Le Roux & Kellogg (1999), with permission from the Botanical Society of America; (b) a pistillate flower, showing undeveloped anthers, in Heteropogon contortus; (c) a staminate flower, showing gynoecium arrested at the gynoecial ridge stage, in Hyparrhenia hirta; (d) adjacent flowers in the inflorescence of H. hirta; the lower flower is in a sessile spikelet and will become bisexual; the upper flower is in a pedicellate spikelet and will become staminate; (e) a staminate flower of Zizania aquatica, showing well-developed gynoecium with ovule and integument. Reproduced from Zaitchik et al. (2000), with permission from the International Journal of Plant Sciences. Bars, 100 µm. an, anther; g, gynoecium; int, integument; lo, lodicule; nu, nucellus; ow, ovary wall.

We hypothesize that there is no single sex-determining pathway in the grasses, but rather that sex expression has been modified in different ways in different lineages. The few studies that have looked at development of unisexual flowers in the grasses support this hypothesis. In staminate flowers of multiple species of panicoid grasses (most of which have staminate and bisexual flowers in the same spikelet), subepidermal cells in the gynoecium undergo cell death soon after formation of the gynoecial ridge (Fig. 1c), a pattern that may characterize the entire clade of approximately 3300 species (Li et al., 1997; Calderon-Urrea & Dellaporta, 1999; Le Roux & Kellogg, 1999; Zaitchik et al., 2000). In staminate flowers in maize, the cells that die in the gynoecium become vacuolized, and lose free ribosomes and other organelles from the cytoplasm (Cheng et al., 1983). However, in wild rice (Zizania aquatica, subfamily Ehrhartoideae, branch 1; Fig. 2), the gynoecium of staminate flowers develops much farther than those in Panicoideae. The stigmatic arms enlarge, the ovule differentiates, and the integuments become visible (Fig. 1e). Growth finally ceases at the stage at which the carpels have nearly closed; developmental arrest of the gynoecium correlates with deposition of dark-staining material in the ovary walls (Zaitchik et al., 2000).

Figure 2.

Best cladogram of relationships of 74 TASSELSEED2 (TS2) and related short-chain dehydrogenase/reductase (SDR) genes from the Bayesian analysis using the General time-reversible (GTR) model, invariant sites and gamma distributed rates (GTR + I + Γ). Bold branches are supported by clade credibility  0.95. B, bisexual flowers; MONO, monocots; P, pistillate flowers; S, staminate flowers; T, TS2/TS2-duplicate clade. 1–5, branches where unisexuality is inferred to have evolved; GxGxxG, NAD- or NADP-binding site; YxxxK, catalysis or subunit interaction motif.

We also hypothesize that development of androecia and gynoecia should be under separate genetic control. Because developmental arrest of the two sorts of organs does not covary in evolutionary time, or even necessarily within the same plant, we conclude that production of a unisexual staminate flower proceeds via a different mechanism from production of a unisexual pistillate flower. They are not, in other words, simply two sides of the same coin.

Genetic basis of dicliny in grasses

The genetic basis of dicliny is best understood in maize, a monoecious plant with staminate flowers borne in apical tassel inflorescences and pistillate flowers in axillary ear inflorescences. In anther ear1 (an1) and dwarf plant1 (d1), d2, d3, d5, and d8 mutants, stamens within the ear florets fail to abort, resulting in a bisexual upper floret and staminate lower floret. Staminate florets in the tassel are unaffected in these mutants. These mutations all affect gibberellin (GA) biosynthesis, indicating that GAs play a key role in the stamen abortion process in ear florets (Phinney, 1961, 1984).

The mutant phenotype of the ear is a reversion to the ancestral state. The genus Zea and its sister Tripsacum were derived from ancestors that had bisexual upper florets and staminate lower florets, the latter pattern being synapomorphic for Panicoideae (Giussani et al., 2001). Thus the origin of one-flowered, pistillate florets in Zea and Tripsacum could have been a response to altered regulation of GA concentrations.

The GA data also confirm that development of different floral organs can be completely decoupled. In the an1 and dwarf mutants, gynoecium development is normal and only anther development is affected.

Multiple loci in maize affect gynoecium development, particularly in the tassel (Coe et al., 1988; Irish et al., 1994). Most of these have pleiotropic effects, affecting aspects of inflorescence branching as well as development of the plant as a whole. However, tasselseed (ts) 1, ts2, and Ts5 mutants, designated class I tasselseed mutants by Irish et al. (1994), affect only gynoecial development in the tassel and lower floret development in the ear, without affecting other aspects of floral, inflorescence or plant development. In ts1, ts2 and Ts5 mutants, sex expression in the tassel is reversed, with the gynoecium developing and stamens being suppressed. In addition, in silkless1 (sk1) mutants the pistils of the ear florets fail to develop, resulting in plants that cannot produce seeds. Staminate florets in the tassel are unaffected in sk1 mutants. Because they affect sex determination specifically and with minimal pleiotropy, TS1, TS2, TS5, and SK1 are good candidates for genes that might have been modified in the evolution of different patterns of sex expression in the grasses.

Double mutant analysis has suggested that TS2 acts in the same pathway as TS1 and TS5 (Irish et al., 1994) and SK1 (Jones, 1934). In ts1; ts2 and ts1; Ts5 double mutants the mutant phenotype resembles the ts1 single mutants, suggesting that TS1 may be upstream of both TS2 and TS5 (Irish et al., 1994). Likewise, ts2; Ts5 double mutants resemble ts2 single mutants, placing TS2 genetically upstream of TS5. In ts2; sk1 double mutants, pistils developed in all ear florets and a mixture of pistillate and staminate florets developed in the tassel. Thus the lack of TS2 gene product suppressed the sk1 mutant phenotype in the ear and lack of SK1 gene product was partially able to correct the ts2 mutant phenotype in the tassel (Jones, 1934; Irish et al., 1994).

Of the class I tasselseed loci, only TS2 has been cloned (DeLong et al., 1993). The gene product resembles a short-chain dehydrogenase/reductase (SDR), with significant similarity to hydroxysteroid dehydrogenases, and is hypothesized to play a role in hormone metabolism (DeLong et al., 1993). Expression of TS2 RNA in would-be staminate flowers correlates with death of cells – presumably the ones in which it is expressed – which lose their cytoplasm; death of these cells presumably prevents continued development of the gynoecium. Calderon-Urrea & Dellaporta (1999) showed that TS2 RNA is expressed in pistils of both pistillate and staminate flowers, although only the pistils of staminate flowers undergo cell death. They hypothesized that the pistil primordium of the pistillate (ear) inflorescence is protected from TS2-mediated cell death by presence of a wild-type SK1 gene, perhaps by formation of a TS2:SK1 protein complex, by disruption of a downstream step in the cell-death pathway, or by sequestration of TS2 or other cell death factors to an organelle compartment (Calderon-Urrea & Dellaporta, 1999). Alternatively, Veit et al. (1993) proposed that absence of TS2 protein in the ts2 mutant resulted in ectopic expression of SK1 in the tassel, leading to feminization. Explicit testing of these hypotheses awaits cloning and characterization of the SK1 gene.

The TS2 ortholog in gama grass (Tripsacum dactyloides, tribe Andropogoneae, subfamily Panicoideae), GYNOMONOECIOUS SEX FORM1 (GSF1), plays a similar role to TS2 in maize (Li et al., 1997). TS2-like SDR genes have also been isolated from A. thaliana (AtATA1) and Silene latifolia (SlSTA1–12), where expression is restricted to tapetal cells, suggesting that these TS2-like SDR genes are not involved in sex determination (Lebel-Hardenack et al., 1997). However, the expression of these genes in tapetal tissue, which breaks down during pollen formation, suggests a more general role in cell death (Lebel-Hardenack et al., 1997). The precise relationship of the A. thaliana and Silene genes to TS2 is not clear, but could easily be determined with additional sequences similar to those of both the dicot and monocot genes.

If TS2 is a major sex-determining protein, then comparative sequencing of TS2 orthologs might be expected to show evidence of positive selection in at least some lineages of grasses. Such tests have only been carried out in two closely related species of Bouteloua. In the dioecious Bouteloua dimorpha, TS2 appears to be evolving neutrally, whereas in the bisexual species Bouteloua hirsuta, TS2 is under strong purifying selection (Kinney et al., 2003). Bouteloua is in the grass subfamily Chloridoideae, and thus is distantly related to subfamily Panicoideae, which includes maize. It is unknown whether gynoecial abortion in Bouteloua correlates with death of cells in the subepidermal layers, and the expression pattern of TS2 is also unknown.

To evaluate the hypothesis that TS2 is a major player in the evolution of unisexual flowers in grasses, we analyzed its molecular evolution and expression in a variety of grasses.

Materials and Methods

Generation and analysis of sequences

Seventy-four TS2 and TS2-like short-chain dehydrogenase/reductase (SDR) genes from 34 eudicots and monocots were examined, including the putative TS2 homologs in Silene latifolia Poir. (SlSTA1–12) and Arabidopsis thaliana (L.) Heynh. (AtATA1 and AtSDR2), and diverse unisexual and bisexual grasses. Forty-one genes were isolated for this study and 33 were identified by blast (Altschul et al., 1997) searches at National Center for Biotechnology Information (NCBI) (http://www.ncbi.nlm.nih.gov/). Degenerate PCR primers to amplify diverse TS2-like SDR genes were designed using primaclade (Gadberry et al., 2005) based on the A. thaliana ATA1, S. latifolia Poir. SlSTA1–12, Tripsacum dactyloides (L.) L. TdGSF1 and Zea mays L. ZmTS2 gene sequences.

Total DNA was isolated using an sodium dodecyl sulphate (SDS) protocol (Dellaporta, 1994), and double-stranded TS2-like SDR PCR products were amplified using a standard reaction mix [2 U Taq polymerase (Promega Corp., Madison, WI, USA), 5 µl of 10 × reaction buffer, 5 µl of 25 mm MgCl2, and 2 µl of 2.5 mm dNTP], plus 20 µm each of the primers TS2-59F, 5′-AGA GGC TGG AMG GGA AGG TG-3′ and TS2–854R, 5′-GTC SAC GAC RAG GTT GTG GC-3′, 10% (by volume) 5 m betaine (Sigma, St Louis, MO, USA), 5% dimethyl sulfoxide (DMSO) (by volume), and 100–200 ng of genomic DNA. PCR reactions used a hot-start, touchdown PCR profile (three cycles at 65°C, three at 63°C, three at 60°C and 25 at 57°C). PCR fragments were purified and subcloned as described by Malcomber & Kellogg (2004), and two to five clones were sequenced per species. Dideoxy sequencing used plasmid primers T7 and SP6, with reactions analyzed on ABI 377, ABI 3100 or ABI 3130XL DNA sequencers (Applied Biosystems, Foster City, CA, USA). Only nucleotide sequences with scores > 20, as determined by PHRED, (Ewing et al., 1998) were used in subsequent analyses. Alignments were edited in Seqman II (DNASTAR Inc., Madison, WI, USA) and all sequences submitted to GenBank (DQ384222–DQ384263).

Nucleotide sequences were aligned based on the conceptual amino acid translation using RevTrans (Wernersson & Pedersen, 2003), and adjusted manually using MacClade 4 (Maddison & Maddison, 2003); nucleotides were used for all analyses. Maximum likelihood (ML) and Bayesian phylogenetic analyses used paup* 4.0 (Swofford, 2000) and MrBayes 3.1 (Huelsenbeck & Ronquist, 2001) on the Beowulf parallel processing cluster at the University of Missouri – St Louis. The ML search used 10 separate heuristic searches with TBR and MULPARS on and 10 random sequence additions. Bayesian analyses used two separate searches of 8 million generations using default flat priors and the General time-reversible (GTR) model of sequence evolution, invariant sites and gamma distributed rates (GTR + I + Γ) (estimated by modeltest; Posada & Crandall, 1998). Trees were sampled every 500 generations and burn-in was determined empirically by plotting likelihood score against generation number. After burn-in trees had been removed, clade credibility (CC) values and the 95% credible set of trees were estimated using MrBayes (Huelsenbeck & Ronquist, 2001).

Copy number of genes was estimated using PCR and confirmed for some taxa by Southern blots. Approximately 10 µg of total DNA was digested with BamHI, EcoRI, or HindIII, separated on a 1.2% agarose gel, blotted onto a nylon membrane, and hybridized with a full-length 32P-dCTP-labeled TS2 probe for 16 h at 65°C following Laurie et al. (1993). After hybridization, blots were washed at 65°C twice in 2 × saline sodium citrate (SSC) (1 × SSC is 0.15 m NaCl and 0.015 m sodium citrate)/0.5% sodium dodecyl sulfate (SDS) for 20 min each time, and twice in 0.1 × SSC/0.1% SDS.

Molecular evolution

Tests for selection on TS2 and related SDR genes used the best ML topology and the codeml program within paml (Yang, 1997) version 3.14 on the Beowulf parallel processing cluster at University of Missouri – St Louis. Evidence of positive selection at particular codons was tested using the nested codon models M0 and M3, M1a and M2a, and M7 and M8 (Yang & Nielsen, 2002; Wong et al., 2004; Yang et al., 2005); the standard likelihood ratio test (LRT) statistic was applied against a χ2 distribution with two degrees of freedom. Model M0 is the simplest codon model with a single ω parameter (ratio of nonsynonymous to synonymous sites, or dN/dS) for all sites and all branches of the phylogeny. M1a is the ‘nearly neutral’ model with two site classes, 0 < ω0 < 1 and ω = 1 (Wong et al., 2004). M2a is the selection model and is an extension of M1a, and has the additional class ω2, which can take any value. M3, the discrete model, is an extension of M0 and has k site classes, each with a separate ω ratio. M3 with two sites classes is the null model for the model B branch-site model described in the following paragraph. M7 and M8 use a discrete β distribution to approximate among-site ω variation. M8 differs from M7 in allowing an additional site class with ω > 1.

We also tested for sites potentially under positive selection on the five branches within the TS2 clade leading to grasses with unisexual flowers (Fig. 2) using the modified branch-site models A and B (Wong et al., 2004; Yang et al., 2005). Model A was compared with model M1a (NearlyNeutral) and model B was compared with M3 (discrete) with two site classes in a LRT against a χ2 distribution with two degrees of freedom (Wong et al., 2004; Yang et al., 2005).

TS2 protein structure modeling

The tertiary structure of the TS2 protein was estimated via homology modeling using the ESyPred3D server (Lambert et al., 2002). The proximity of sites that were identified as being potentially under positive selection to the NAD/NADP binding site (GxxGxG) and catalysis or subunit interaction site (YxxxK) was investigated using protein explorer 2.75 (Martz, 2002).

Gene expression

Total RNA was extracted from developing rice inflorescences, culms, young leaves and roots using RNAwiz solution (Ambion, Austin, TX, USA) according to the manufacturer's instructions. TS2 expression profiles were inferred using the Superscript One-step RT-PCR kit (Invitrogen, Carlsbad, CA, USA) following the manufacturer's instructions, except that 4%[volume/volume (v/v)] DMSO and 10% (v/v) 5 m betaine (Sigma) were included to facilitate strand separation. TS2-specific PCR primers were designed using primaclade (Gadberry et al., 2005). Oryza sativa L. TS2 (OsTS2) gene fragments were amplified with a 55°C annealing temperature using the primers OsTS2-828F, 5′-GAA GAT GGA GGA GGT GGT CA-3′, and OsTS2-1110R, 5′-AGT CCA ATT AAC ACA TTG AAT CAA GA-3′. Zea mays TS2 (ZmTS2) and Sorghum bicolor (L.) Moench TS2 (SbTS2) gene fragments were both amplified with a 57°C annealing temperature using the primers ZmTS2-915F, 5′-GAG GTG GAG AAG ATG GAG GAG-3′ and ZmTS2-1196R, 5′-ATG AAT CAA TCA ACC AAA TGA AAA-3′ (for maize), and SbTS2-43F, 5′-ATC GTG MGK CTG TTC GTG A-3′, and TS2-854R (for sorghum). All RT-PCR products were verified by subcloning and sequencing as previously described. The amplified TS2 fragment lacks introns, so that the cDNA and gDNA PCR products are expected to be the same size. We therefore included triose phosphate isomerase (TPI), which has introns, as a positive control to verify that none of the RNA extractions was contaminated by gDNA. Exons four and five of TPI were PCR-amplified using the same conditions using degenerate primers TPIX4F, 5′-AAG GTC ATT GCA TGT GTT GG-3′, and TPIX6R, 5′-TTT ACC AGT TCC AAT AGC CA-3′ (Strand et al., 1997), which span a 500–800-bp intron. If RNA were contaminated with DNA we would see two bands in the RT-PCR reaction.

RNA in situ hybridization was conducted on developing rice inflorescences using 3′ untranslated region (UTR) probes derived from RT-PCR products using the OsTS2-828F and OsTS2-1110R primers, as described in Malcomber & Kellogg (2004).


TS2 is single-copy in the grasses, but has an ancient duplicate

PCR fragments ranged from 710 bp (Flagellaria indica L.) to 804 bp (Hordeum vulgare L.) and represent ∼79% of the Z. mays TS2 coding region. Alignment of the predicted amino acid sequences was largely unambiguous, except for a hypervariable region between amino acids 225 and 242 that was removed from subsequent analyses (Fig. 3). Interestingly, this region represents an apparent insertion in members of the Bambusoideae, Ehrhartoideae, and Pooideae, which together form a clade (the BEP clade) in some phylogenetic analyses (Grass Phylogeny Working Group, 2001).

Figure 3.

C-terminal amino acid alignment of 74 TASSELSEED2 (TS2)-like SDR proteins included in the analysis, with the hypervariable region between amino acids 225 and 242 that was removed from subsequent analyses shaded gray.

The best ML tree for the 75 SDR gene dataset (–ln 17592.537) was identical to the tree with the highest posterior probability from the Bayesian analysis, and estimated a well-supported ( 0.95 clade credibility) clade of monocot SDR genes, and several well-supported subclades (Fig. 2). The 95% credible set of topologies included 15969 distinct tree topologies, indicating that the SDR dataset has only limited resolving power.

The Z. mays TS2 (ZmTS2) sequence falls within a well-supported clade that we will refer to as the TS2/TS2-duplicate clade (T; Fig. 2). The TS2/TS2-duplicate clade contains a broad sample of grasses, including a representative of the earliest diverging lineage, Anomochloa, indicating that the duplication event producing the two lineages occurred before the origin of extant grasses. Sister to the TS2/TS2-duplicate clade is a well-supported clade of SDR genes from several grasses and F. indica, a member of the Poales. The presence of 10 distinct rice sequences within this clade indicates that multiple duplication events occurred during diversification of this lineage, although current sampling limits our ability to infer where within monocots the duplication events occurred. The A. thaliana genes At2g47140 and At2g47130 are inferred to be sister to the clade of monocot SDR genes, although this relationship is not well supported. Other closely related A. thaliana SDR genes include At3g51680 (= AtSDR2; Cheng et al., 2002) and AtATA1 (Fig. 2). However, the complex pattern of duplication of SDR genes within monocots means that none of the A. thaliana genes is orthologous to ZmTS2.

Estimated relationships within the TS2 clade by the Bayesian and ML analyses are largely congruent with the Grass Phylogeny Working Group (2001) analysis, although the placement of Pharus and Anomochloa is reversed in the TS2 phylogeny relative to the inferred organismal phylogeny. However, this placement in the TS2 topology is not well supported.

We isolated up to two distinct TS2 sequences per species (A-B; Fig. 2) from sequencing two to five clones. Kinney et al. (2003) reported up to 20 TS2 alleles in their analysis of Bouteloua dimorpha J.T. Columbus and Bouteloua hirsuta Lag. All TS2 sequences from the same species coalesce in our analysis except for B. hirsuta (Fig. 2), with raw divergence among clones ranging from 0.005 (Ehrharta erecta Lam.) to 0.118 (Brachypodium distachyon (L.) P. Beauv.). Both sequenced clones in B. distachyon have an open reading frame, suggesting that they are functional. B. distachyon (n = 5) may be diploid, tetraploid, or hexaploid (Index of Plant Chromosome Numbers; http://mobot.mobot.org/W3T/Search/ipcn.html), indicating that the high divergence between the sequenced colonies may reflect different TS2 copies within our polyploid accession. The minimal variation detected among TS2 clones in the other species is likely caused by PCR error, although some of the variation could be caused by persistence of ancestral polymorphism (lineage sorting). One B. hirsuta clone forms a clade with Bouteloua trifida Thurb. ex S. Watson which is sister to a clade of TS2 sequences from diverse chloridoid grasses (Bouteloua, Eragrostis, Muhlenbergia, and Spartina), suggesting a duplication near the base of subfamily Chloridoideae or possibly deeper within the PACCAD clade.

We verified the copy number of the TS2-like genes in grasses by Southern blots using species-specific TS2 probes in the diploid species O. sativa, Pharus lappulaceus Aubl. and S. bicolor (Index of Plant Chromosome Numbers; http://mobot.mobot.org/W3T/Search/ipcn.html). Only one gene was detected on the O. sativa and S. bicolor blots, whereas a faint second copy was visible on the P. lappulaceus blot (Fig. 4). Uncorrected sequence divergence between the rice TS2 and TS2-duplicate genes was 0.206, so the TS2 probe was unlikely to hybridize to the TS2-duplicate using our high-stringency hybridization conditions. In contrast, the faint extra band in the HindIII lane of the P. lappulaceus blot could represent the TS2-duplicate gene. Sequence divergence between TS2 and the TS2-duplicate gene in Anomochloa, the other early diverging grass included within our analysis, was 0.18207, just within the limits of detection on a high-stringency DNA gel blot. HindIII does not cut within the P. lappulaceus TS2 gene fragment and the low sequence divergence between the two distinct P. lappulaceus TS2 clones (0.0078) included in our analysis is more likely to have been a result of Taq error than different gene copies.

Figure 4.

DNA gel blots of Oryza sativa, Pharus lappulaceus and Sorghum bicolor. Blots were probed with species-specific TASSELSEED2 (TS2) probes. Genomic DNA was digested with BamHI (B), EcoRI (E) and HindIII (H) restriction enzymes.

The sequence of TS2 is highly conserved and under strong purifying selection

All SDR enzymes have an N-terminal NAD- or NADP coenzyme-binding pattern of GxxGxG and a motif considered to be involved in catalysis or subunit interaction of YxxxK (Jörnvall et al., 1999). The GxxGxG motif is conserved as GARGIG in all TS2 and TS2-like sequences, except that the O. sativa TS2-duplicate and Orthoclada laxa (Rich.) P. Beauv. TS2 and TS2-duplicate sequences have Ser instead of Ala, the Danthonia spicata (L.) P. Beauv. ex Roem. & Schult. TS2-duplicate sequence has Gln instead of Arg, and the Anomochloa marantoidea Brongn. TS2-duplicate sequence has Gly instead of Arg (C to G at position 1). The YxxxK motif is conserved as YTASK in all TS2 and TS2-duplicate sequences except that A. marantoidea TS2A and O. laxa TS2-duplicate both have Val instead of Ala (a conservative substitution).

Despite widespread conservation of the GxxGxG binding site in all other SDR genes (Jörnvall et al., 1999), several monocot SDR genes have an Ala instead of a Gly at the first position of the NAD/NADP binding domain (Fig. 2). This amino acid change might affect their ability to bind NAD/NADP, although biochemical analyses have not been conducted on these proteins. In monocot SDR genes where the GxxGxG and YxxxK motifs are conserved, they are generally not GARGIG and YTASK, also indicating possible diversification of biochemical function.

All site-specific models indicate that TS2 sequences were under strong purifying selection (ω << 1.0) with no evidence for positive selection having acted on sites across the whole phylogeny. The one-ratio model (M0) estimated ω = 0.0934. The discrete model (M3) was significantly better than the one-ratio model (P < 0.001), indicating considerable variability in ω among sites, but also identified no positively selected sites. The nearly neutral model (model M1a) allowed two ratios, 0 < ω0 < 1 and ω1 = 1, and estimated 97.1% of sites to be under considerable purifying selection (ω0 = 0.029). The more complex models also failed to find any evidence of positive, or even relaxed, selection.

Using branch-site models A and B, we tested for evidence of positive section on the branches where unisexuality is inferred to have evolved. Our species sample included five such branches: (1) Zizania aquatica (L.), (2) the herbaceous bamboo clade (Lithachne, Olyra and Pariana), (3) B. dimorpha, (4) the panicoid clade (Ixophorus, Setaria, Sorghum, Tripsacum and Zea), and (5) P. lappulaceus (1–5; Figs 2 and 5). Testing all five unisexual branches simultaneously, we found ω2 greater than 1 in model A (ω2 = 1.796), in which two sites were identified as potentially being under positive selection (Ala29 to Leu and Arg117 to Gly); however, neither site had a significant posterior probability (0.507 and 0.503, respectively) and the model A likelihood score was not significantly better than that of the null model. Positive selection was also inferred in model B (ω2 = 1.418), and the likelihood score was significantly better than that of the null model, but no sites were identified as potentially under positive selection.

Figure 5.

Maximum likelihood (ML) phylogram of the TASSELSEED2 (TS2) clade with nonsynonymous changes on each of the five branches (1–5) where unisexuality is inferred to have evolved as estimated by paml (Yang, 1997). The number represents the amino acid position within the TS2 protein followed by the estimated change and the likelihood in parentheses.

Tests for positive selection on each branch individually identified the branch subtending the herbaceous bamboo clade (Lithachne, Otatea and Pariana) (branch 2; Fig. 2). Under both models (A and B) the foreground parameter ω2 was large and likelihood scores were significantly better than that of the null model. However, only in model A was a site potentially under positive selection identified with a significant posterior probability, with replacement of Asp32 (D) by Gln (Q) (P = 0.972). ω was also greater than 1 in analyses of the branch subtending the Pharus clade (branch 5), but only model B was significantly better than the site model and failed to identify any sites as being potentially under positive selection (Table 1). Model A identified two sites as being potentially under positive selection (Ala29 to Leu and Leu86 to Ala), but neither change had posterior probability  0.95 and the likelihood score for the branch site model was not significantly better than that for the site model (Table 1).

Table 1.  Codon model parameter estimates for TASSELSEED2 (TS2)
Modelpln2Δ lnParameter estimates
  • Bayes empirical Bayes (BEB) tests detected no positively selected sites at > 0.95 posterior probability, except for branch site model A for the herbaceous bamboo lineage (branch 4; see text for details).

  • *

    P < 0.05;

  • **

    P < 0.01.

M0 (one ratio)1−19206.608 ω = 0.0934
Site models
M1a (nearly neutral)2−19170.500 p0 = 0.971 (p1 = 0.029)
   ω0 = 0.1259
M2a (positive selection)4−19170.500  0p0 = 0.968, p1 = 0.032 (p2 = 0)
   ω0 = 0.1031, ω2 = 25.997
M3 (discrete, K = 2)3−18802.531808.154**p0 = 0.366, p1 = 0.634
   ω0 = 0.021, ω1 = 0.148
M7 (beta)2−18712.816 P = 0.897, q = 7.843
M8 (beta and ω)4−18712.816  0p0 = 1.00, p = 0.897, q = 7.843
   (p1 = 0.00) ω = 3.067
Branch site models
Foreground 1–5
Model A4−19170.373  0.254p0 = 0.966, p1 = 0.029 (p2 + p3 = 0.005)
   ω0 = 0.099, ω2 = 1.796
Model B5−18799.271  6.62*p0 = 0.347, p1 = 0.636 (p2 + p3 = 0.0167)
   ω0 = 0.019, ω1 = 0.146, ω2 = 1.418
Foreground 1
(Zizania aquatica)
Model A4−19170.500  0p0 = 0.971, p1 = 0.029 (p2 + p3 = 0)
   ω0 = 0.100
Model B5−18800.470  4.122p0 = 0.113, p1 = 0.205 (p2 + p3 = 0.6817)
   ω0 = 0.020, ω1 = 0.143, ω2 = 0
Foreground 2
(herbaceous bamboo clade)
Model A4−19165.557  9.886**p0 = 0.962, p1 = 0.029 (p2 + p3 = 0.009)
   ω0 = 0.099, ω2 = 999
Model B5−18795.018 15.026**p0 = 0.353, p1 = 0.639 (p2 + p3 = 0.008)
   ω0 = 0.019, ω1 = 0.141, ω2 = 999.00
Foreground 3
(Bouteloua dimorpha)
Model A4−19170.450  0.1p0 = 0.970, p1 = 0.029 (p2 + p3 = 0)
   ω0 = 0.100
Model B5−18800.743  3.576p0 = 0, p1 = 0 (p2 + p3 = 1)
   ω0 = 0.020, ω1 = 0.142, ω2 = 0
Foreground 4
(Panicoid clade)
Model A4−19170.450  0.1p0 = 0.971, p1 = 0.029 (p2 + p3 = 0)
   ω0 = 0.100
Model B5−18802.360  0.342p0 = 0.234, p1 = 0.424 (p2 + p3 = 0.342)
   ω0 = 0.020, ω1 = 0.142, ω2 = 0
Foreground 5
(Pharus lappulaceous)
Model A4−19169.685  1.63p0 = 0.962, p1 = 0.029 (p2 + p3 = 0.009)
   ω0 = 0.099, ω2 = 6.797
Model B5−18799.351  6.36*p0 = 0.348, p1 = 0.643 (p2 + p3 = 0.008)
   ω0 = 0.019, ω1 = 0.140, ω2 = 6.41

The Asp32 to Gln substitution on the branch subtending the herbaceous bambusoid clade requires substitutions at the first and third positions of the codon, and changes the charge of the residue. The tertiary structure of the Lithachne humilis Soderstrom TS2 protein shows that position 32 is not near either known active site (Fig. 6), although the Asp to Gln change could affect other as yet unidentified active or binding sites within the protein.

Figure 6.

Tertiary structure of the Lithachne humilis TASSELSEED2 (TS2) protein with NAD, NADP or NADPH (GxGxxG), subunit interaction or catalysis (YxxxK), inferred ligand position, and site potentially under positive selection indicated.

Other amino acid changes occur in parallel, and do not correlate with unisexuality. Ala66 changed to Ser along the unisexual panicoid branch, on the branch subtending the pooid clade (Avena, Brachypodium, Psathyrostachys and Melica) and on the branch leading to the chloridoid grass Muhlenbergia. Similarly, three nonsynonymous changes were estimated on the Zea/Tripsacum branch: Val20 to Ala, Leu112 to Arg, and Iso162 to Val. Identical changes at positions 20 and 112 occur on the branch subtending the B. hirsuta clade, and the change from Iso162 to Val was also inferred for the branches subtending the Muhlenbergia, Spartina and Anomochloa TS2 sequences. All these changes are located on the periphery of the TS2 protein, and none appears to affect either of the known active sites.

TS2 expression is not restricted to flowers

Models of TS2 function imply that its primary role should be in flowers, and predict that it should be expressed primarily in inflorescences. We were unable to find any reports in the literature, however, that tested this assumption, and therefore performed the relevant RT-PCR with TS2-specific primers on inflorescence, culm, leaf and root RNA from O. sativa, S. bicolor and Z. mays. Oryza sativa has spikelets with one flower bisexual and fertile, and two strongly reduced and sterile; the latter do not initiate either gynoecium or androecium. If the primary role of TS2 is to suppress gynoecium development, we would expect no expression in O. sativa. Sorghum bicolor has paired pedicellate and sessile spikelets, each with two flowers; the lower flower is always sterile and, like sterile flowers in rice, does not initiate floral organs. The upper floret of the sessile spikelet is bisexual, whereas that of the pedicellate spikelet is staminate or sterile. We expected some TS2 expression in S. bicolor, consistent with suppression of the gynoecium in the upper flower of the pedicellate spikelet. As discussed above, Z. mays is monecious with staminate flowers borne in apical tassel inflorescences and pistillate flowers borne in axillary ear inflorescences. We expected TS2 expression in both inflorescences, based on the mutant phenotypes.

Surprisingly, TS2 is expressed in inflorescence, leaf, culm, and root tissue of all three species (Fig. 7). Amplification of TPI, which unlike TS2 has introns, verified that none of the RNA extractions were contaminated by gDNA.

Figure 7.

TASSELSEED2 (TS2) RNA RT-PCR expression patterns in inflorescences (Infl.), leaf, culm, and root tissues of Oryza sativa (Os), Sorghum bicolor (Sb) and Zea mays (Zm). Genomic DNA (gDNA) and TRIOSE PHOSPHATE ISOMERASE (TPI) RT-PCR were included as a positive control and to check for DNA contamination.

To investigate the expression pattern of TS2 in the gynoecium of a species with bisexual florets, we conducted RNA in situ hybridization on developing rice flowers. The gene is expressed in the developing pistil and in anthers (Fig. 8). The pistil is functional in rice, indicating that, as in maize, TS2 expression alone does not indicate which organs will undergo programmed cell death. Stamens in rice flowers appear to express TS2 throughout development, and expression is particularly apparent in mature anthers (Fig. 8b).

Figure 8.

In situ hybridization of TASSELSEED2 (TS2) RNA expression in developing spikelets of Oryza sativa with antisense (a, b) and sense (c) digoxygenin-labeled RNA probes. (a, b) Developing spikelet with expression in anther (an), pistil (pi) and palea (pa). (c) Control section showing lack of hybridization signal using the sense probe. Bars, 100 µm.


Sex expression has changed many times in the evolution of the grass family, and in disparate ways. Discussion of ‘unisexual flowers’ is thus somewhat misleading. The morphological diversity and repeated origins of unisexual flowers make it unlikely that there is a single master switch that affects transitions from unisexual to bisexual or vice versa. The variability of sex expression in the grasses suggests instead that multiple genes may be involved, and that modification of any one of them could lead to a different pattern of floral organ development.

Comparative morphology shows that androecial development is commonly decoupled from gynoecial development. For example, Ixophorus unisetus has two flowered spikelets, with the upper flower pistillate and lower one staminate (Kellogg et al., 2004). The species is derived from ancestors in which the upper flower was bisexual; in the history of Ixophorus only stamen development in one flower of the spikelet was altered. In Hyparrhenia hirta, the upper flower of all pedicellate spikelets is staminate, but the upper flower of the sessile spikelets varies in sex expression depending on position in the inflorescence; sessile spikelets in the lower part of the inflorescence are staminate, whereas those in the upper part of the inflorescence are bisexual (Clayton & Renvoize, 1986). Space does not permit an exhaustive list of the possible combinations of staminate, pistillate, sterile, and bisexual flowers, but the many possibilities indicate precise genetic control and presumably ease of modification.

Stamen development is clearly controlled, at least in part, by GA. Blocking GA production in maize ear florets stops stamen abortion, producing the bisexual ear florets seen in an1 and several d mutants (Phinney, 1961, 1984; Dellaporta & Calderon-Urrea, 1994). Stamen abortion in tassel florets of ts2 mutants is also ascribed to perturbation of the GA pathway (Dellaporta & Calderon-Urrea, 1994), although this has not been rigorously tested. This suggests that formation of pistillate flowers may reflect modifications in availability of or sensitivity to GA during flower development.

Cytokinin is also apparently involved in regulating floral organ development. Ectopic expression of the cytokinin-synthesizing isopentyl transferase (IPT) permits the pistil in the lower floret in maize ear spikelets to develop, producing plants with two functional pistils in each spikelet (Young et al., 2004). Curiously, the IPT transgene had no effect on floral development or sex expression in the tassel.

Gynoecial development is affected by TS1, TS2, TS5, and SK1, but the cellular and biochemical function of these proteins is unknown and only TS2 has been cloned. Although TS2 was cloned over a decade ago, little information is available in the literature, perhaps in part because it is a difficult gene to work with. The GC content is high (∼72%), and PCR amplification is unreliable without additives that reduce secondary structure such as DMSO and betaine, even with nondegenerate primers.

TS2 is a member of the large SDR protein superfamily, which comprises over 3000 members in over 1000 forms that act on diverse substrates including alcohols, sugars, steroids and aromatic compounds (Kallberg et al., 2002). NAD, NADP or NADPH (GxxGxG) and subunit interaction or catalysis (YxxxK) motifs are conserved among all SDRs; the precise sequence of each motif is conserved among orthologs and differs among paralogs. Within the TS2 clade, these motifs are GA/SRGIG and YTASK. The substrate is unknown, but its high sequence similarity to hydroxysteroid dehydrogenases suggests a gibberellin or steroid-like molecule (Calderon-Urrea & Dellaporta, 1999). The SDRs characterized in A. thaliana (AtATA1 and AtSDR2) and S. latifolia (STA1–12) are not orthologous to any known monocot SDRs, including TS2 itself, so function cannot be directly extrapolated from those two eudicots to the grass gene products. Functional studies will need to be conducted in the grasses themselves.

TS2 is expressed throughout the plant in maize, rice, and sorghum. It is expressed in both gynoecium and androecium in rice. In maize, it is expressed in all gynoecia, whether they develop fully or not (DeLong et al., 1993; Calderon-Urrea & Dellaporta, 1999). We suggest that TS2 probably has a general developmental role, and that its function in sex determination is ancillary and perhaps taxonomically restricted.

TS2 is necessary for abortion (i.e. normal development) of the gynoecium in staminate flowers of maize and Tripsacum (DeLong et al., 1993; Li et al., 1997). Expression of TS2 in would-be staminate flowers correlates with death of cells – presumably the ones in which it is expressed – which lose their cytoplasm; death of these cells presumably prevents continued development of the gynoecium. A similar pattern of gynoecial cell death in staminate flower development is seen in other panicoid grasses (Le Roux & Kellogg, 1999).

If TS2 had been recruited multiple times for specification of unisexual flowers, we might expect to find an elevated rate of amino acid replacements in one or more lineages. However, most data point to extensive purifying selection, indicating strong conservation of the protein sequence. Most of the amino acid replacements that do occur are conservative, occur more than once in the evolution of the grasses, and do not correlate with the origin of unisexual flowers. Only along the branch leading to the herbaceous bamboo lineage do we find evidence of selection on a single site, a nonconservative D to Q substitution.

In the case of the dioecious B. dimorpha clade, the origin of unisexual flowers does not correlate with any nonsynonymous changes at all. Kinney et al. (2003) examined the molecular evolution of 18 TS2 alleles in B. dimorpha and concluded that the locus was evolving neutrally, which would suggest an elevated rate of nonsynonymous mutations. However, they used a population genetic test (Tajima's D statistic), rather than the likelihood ratio tests that directly address codon changes as reported here; they also amplified a longer TS2 fragment so additional residues were certainly examined by their test. It is not therefore clear whether our results actually conflict.

Widespread expression of TS2, strong conservation of sequence, and lack of correlation of sequence variation and expression with unisexuality all suggest that TS2 is in fact not a sex determination protein. It seems more likely that TS2 may have been co-opted for sex determination in one lineage of grasses (subfamily Panicoideae) but that it has multiple other roles. Additional insight may come from investigating the possible regulators of TS2, which have been identified by genetic, microarray, and bioinformatics approaches. At the same time, we should note that we find broad expression patterns and strong purifying selection on TS2 even in Zea/Tripsacum, where the protein is believed to operate in sex expression. We therefore cannot completely rule out the possibility that it is involved in other lineages, but that we have no means to detect it.

TS2 expression is transcriptionally regulated by hormone levels, as shown by microarray experiments. Rice TS2 is up-regulated in callus tissue exposed to GA and down-regulated when exposed to abscisic acid (ABA), indicating that the gene may play a role in the interchange between these two pathways (NCBI Gene Expression Omnibus accession GPL477; Yazaki et al., 2003). The fact that TS2 is expressed in callus tissue at all is evidence that its role is more general than determination of sex expression in flowers. Although usually regarded as a stress hormone and growth inhibitor, ABA in A. thaliana also plays a key role in controlling fertility, promoting vegetative growth and determining organ size (Cheng et al., 2002). Whether ABA functions in a similar way in grasses, and whether ABA and GA factor into TS2-mediated gynoecial cell death in panicoid grasses, is unknown.

The widespread expression of TS2 RNA that we have observed points to possible post-transcriptional regulation. miRNAs within the 3′ UTR could block translation, as in APETALA 2 (AP2) in A. thaliana (Aukerman & Sakai, 2003) and as hypothesized for lineage (lin)-4 in Caenorhabditis elegans (Bartel, 2004). Based on the output of MicroInspector (Rusinov et al., 2005), the 3′ UTR of rice TS2 has binding sites for miR164 and miR419, whereas maize TS2 has sites for miR169, miR172, miR399, miR439 and miR440. Little is known about the role of these microRNAs, although miR164 is known to regulate hormone response and miR172 regulates translation (Kidner & Martienssen, 2005). microRNA expression profiles and TS2 protein levels could be compared to test whether TS2 translation is regulated by microRNAs.

TS2 and the origin of unisexual flowers

Our data illustrate the power of a comparative approach for developing and testing hypotheses of gene function. At the same time, we show the perils of trying to explain the origin of a complex phenotype by studying the evolution of a single gene. Our data suggest that the role of TS2 is much broader than simply killing cells in the gynoecium of maize, that it may be regulated post-transcriptionally as well as transcriptionally, and that the protein sequence is highly conserved among grasses.

When we began this study, we postulated that TS2 might function in staminate flower specification in many lineages with unisexual flowers, but not in development of bisexual flowers. Expression data in rice, maize, and sorghum, and sequence data from multiple grasses all argue against this hypothesis. The protein sequence is conserved among grasses, indicating that any modifications of its developmental role are likely to be regulatory. The gene is expressed throughout the plant in the three disparate grasses investigated, indicating a function much more general than simple specification of gynoecial development. The gene is expressed in gynoecia that do develop fully as well as those that abort, indicating that even in gynoecial development it does not function as a simple on/off switch. We therefore postulate that TS2 may function in gynoecial development in all grasses, and its role in gynoecial abortion in the Zea/Tripsacum clade is a result of alterations in the activity or regulation of other genes in the gynoecial development pathway.

TS2 is regulated positively by TS1 and apparently negatively by SK1, neither of which has been cloned. One tantalizing possibility is that SK1 is a microRNA that blocks translation of TS2. Only if the microRNA were absent would the TS2 protein be present and functional.

Studies of the genetics of floral organ development in species other than maize will ultimately be necessary. The placement of staminate and pistillate flowers in separate inflorescences in maize adds a complication to interpretation of results, in that the fate of androecia or gynoecia in the tassel is not always the same as that of androecia or gynoecia in the ear. It would be of considerable interest to study floral organ mutations in species in which both sexes are in the same inflorescence. Fortunately, increasing numbers of genomic and genetic tools are available for multiple species of grasses. We can hope that over the next several years more detailed functional hypotheses will be available to test.


We thank the members of the Kellogg lab and three anonymous reviewers for comments and suggestions, Lynn Clark for insights into relationships and morphology of bamboos and leaf material of Anomochloa and Flagellaria, and Jessie Kossuth for help with sequencing TS2 clones. This research was supported by grants from the University of Missouri, the National Science Foundation, and the E. Desmond Lee and Family Foundation.