•Hardened floral bracts and modifications to the inflorescence axis of grasses have been hypothesized to protect seeds from predation and/or aid seed dispersal, and have evolved multiple times independently within the family. Previous studies have demonstrated that mutations in the maize (Zea mays ssp. mays) gene teosinte glume architecture (tga1) underlie a reduction in hardened structures, yielding free fruits that are easy to harvest. It remains unclear whether the causative mutation(s) occurred in the cis-regulatory or protein-coding regions of tga1, and whether similar mutations in TGA1-like genes can explain variation in the dispersal unit in related grasses.
•To address these questions TGA1-like genes were cloned and sequenced from a number of grasses and analyzed phylogenetically in relation to morphology; protein expression was investigated by immunolocalization.
•TGA1-like proteins were expressed throughout the spikelet in the early development of all grasses, and throughout the flower of the grass relative Joinvillea. Later in development, expression patterns differed between Tripsacum dactyloides, maize and teosinte (Z. mays ssp. parviglumis).
•These results suggest an ancestral role for TGA1-like genes in early spikelet development, but do not support the hypothesis that TGA1-like genes have been repeatedly modified to affect glume and inflorescence axis diversification.
Flowering plants have evolved a tremendous number of strategies to increase fitness through modification of their inflorescence and floral structures. The grass family has particularly complex inflorescences and structures surrounding the flowers (Cheng et al., 1983; Clifford, 1987; Ikeda et al., 2004). Each flower (floret) is made up of a gynoecium, androecium, modified inner tepals (lodicules), an adaxial bract (palea) and an abaxial bract (lemma). The flowers are then aggregated into short spikes (spikelets), which are subtended by two more bracts (glumes). Although the gynoecium and androecium are very similar in most grasses, the structure of the paleas, lemmas and glumes, and the arrangement of the spikelets in the inflorescence, all vary extensively among the 12 000 species of grasses.
A few grass species have independently evolved an inflorescence structure in which the spikelets are embedded in the inflorescence axis (rachis). In these species, the axis develops so that it surrounds the developing spikelet, forming a cup-like depression (Clayton & Renvoize, 1986; E. A. Kellogg, pers. obs.). One or both glumes then cover the outside of the spikelet, similar to a trap door hinged at the bottom; the glumes spread outwards from the inflorescence axis at anthesis and then close again after pollination. In inflorescences such as this, the glumes are often leathery or hardened.
Inflorescences with embedded spikelets have originated in several of the grass subfamilies, but are particularly common in the tribe Andropogoneae, subfamily Panicoideae (Clayton & Renvoize, 1986; E. A. Kellogg, pers. obs.). In most members of Andropogoneae, the rachis of the inflorescence breaks apart (disarticulates) at the node below each spikelet, such that the dispersal unit includes the spikelet plus its adjacent internode. In species in which the spikelet is embedded in the rachis, and in which the glumes are hardened, the mature fruit is fully enclosed in a case made up of the rachis internode plus the glume(s). This structure has been postulated to protect the reproductive structures from damage and/or to facilitate seed dispersal (Wilkes, 1967), although its selective value has never been tested. Despite the important ecological and economic implications for such structures, little is known about how their development is controlled at the genetic level.
The best-studied species with spikelets embedded in the rachis and covered with hardened glumes is teosinte in the genus Zea. In Zea (tribe Andropogoneae, subfamily Panicoideae), variation in the SQUAMOSA PROMOTER BINDING PROTEIN-LIKE (SPL) gene, teosinte glume architecture (tga1), underlies a major quantitative trait locus (QTL) for the domestication of maize (Zea mays ssp. mays) from its progenitor teosinte (Z. mays ssp. parviglumis) (Dorweiler et al., 1993; Wang et al., 2005). SPL genes have a wide variety of developmental roles in angiosperms – including the regulation of vegetative and inflorescence phase change, vegetative and inflorescence branching and fruit development – and are known to target the regulation of SQUAMOSA/FRUITFULL (SQUA/FUL)-like genes, commonly involved in inflorescence and flower development (Mandel & Yanofsky, 1995; Klein et al., 1996; Manning et al., 2006; Schwarz et al., 2008; Wang et al., 2009; Jiao et al., 2010; Preston & Hileman, 2010).
Phenotypic differences associated with the maize tga1 QTL determine the ease with which the fruit can be separated from the surrounding inflorescence structures. Female inflorescences (ears) of teosinte have remarkably hard glumes with high levels of silica deposition and a high ratio of small to large cells in the mesophyll at maturity, deeply invaginated rachis internodes and little elongation of the floral branch (rachilla). Conversely, maize ears have glumes with fewer small cells, more lignin and less silica deposition, little invagination of the rachis and greater elongation of the rachilla (Clayton & Renvoize, 1986; Dorweiler & Doebley, 1997). Introgression of the maize allele into a teosinte background results in less internode invagination, such that the fruit is no longer enclosed by the rachis. Conversely, introgression of the teosinte allele into a maize background causes increased internode invagination and thickening of the outer glume (Wang et al., 2005). These differences suggest that tga1 is important for controlling both the hardness of the glumes and the growth of the inflorescence axis to enclose the fruit.
The tga1 alleles of teosinte and maize differ by only seven nucleotides (Wang et al., 2005). One of these differences encodes a nonconservative amino acid substitution from lysine to asparagine at position six in the protein, and has been hypothesized to affect protein stability or function. This is supported by the teosinte-like phenotype resulting from ethyl methanesulfonate mutagenesis of maize, causing a nonconservative mutation at a neighboring amino acid (position five) (Wang et al., 2005). The remaining six changes are located in the promoter region, and have been hypothesized to affect gene regulation. In early- to mid-stage female spikelets, tga1 mRNA transcript levels are equivalent between maize and teosinte (Wang et al., 2005). The gene is expressed in the floret meristem, gynoecium, lodicules, paleas, lemmas and glumes. By contrast, at the same time points, levels of protein accumulation, as measured by western blots, appear to be markedly different between the two subspecies (Wang et al., 2005). Protein levels are significantly higher in both early and mid-stages of ear development in teosinte and in maize lines carrying the teosinte tga1 allele (Wang et al., 2005).
The available data suggest that changes to TGA1 protein stability, translational efficiency or function are more probable explanations for the maize phenotype than are changes in the regulation of the tga1 gene (Wang et al., 2005). Thus, the lysine to asparagine amino acid substitution may be one of the causative sites underlying the morphological diversification of the maize inflorescence. This amino acid change hypothesis predicts that patterns of mRNA expression will be similar between teosinte and maize post-pollination (> 22 mm ear or style elongation (silk) stage), the stage at which hardening of the outer glume and invagination of the rachis occur in teosinte (Dorweiler & Doebley, 1997). Furthermore, if changes in TGA1 protein stability or translational efficiency, rather than protein function, underlie phenotypic differences, distinct patterns of protein accumulation can be predicted between the two.
The developmental role of tga1 must be more complex than just described for the female inflorescences of maize and teosinte. Although the female spikelets of teosinte are embedded in the rachis and have rock-hard glumes, the male (tassel) spikelets are not embedded and the glumes are firm, but leaf-like. Furthermore, the TGA1-like sequences of two distantly related grasses, rice (Oryza sativa) and wheat (Triticum aestivum), share the lysine residue of teosinte at position six (Wang et al., 2005), suggesting that this is the ancestral state for most grasses. If this residue were perfectly correlated with glume and inflorescence phenotype, rice and wheat should have hard glumes and embedded spikelets, like teosinte. However, the glumes of wheat are leaf-like (membranous to coriaceous), the glumes of rice are tiny and flap-like and neither species has spikelets embedded in the inflorescence axis; in this respect, they are similar to most other grasses. These observations suggest that different developmental pathways are responsible for glume and rachis architecture in male and female teosinte inflorescences, and that the role of TGA1 in glume and rachis architecture may have diverged more than once within the tribe Andropogoneae.
The sister genus of Zea is Tripsacum (Lukens & Doebley, 2001; Mathews et al., 2002; Bomblies & Doebley, 2005), which is, like Zea, monoecious. Female spikelets of Tripsacum have thick hardened glumes that are sunken into the equally hard rachis, although neither the glumes nor the rachis is as solid as those of teosinte. Male spikelets of Tripsacum have leathery glumes that are somewhat firmer than those of Zea, but much more flexible than the glumes of the female spikelets. Other Andropogoneae species with invaginated internodes and thickened glumes include Rhytachne, Coelorachis and various other genera in the tribe (sensuClayton & Renvoize, 1986; Mathews et al., 2002). Unfortunately, previous phylogenetic analyses have found little support for relationships within Andropogoneae (e.g. Lukens & Doebley, 2001; Mathews et al., 2002), making the evolutionary history of glume hardening and rachis internode invagination difficult to reconstruct.
In this study, we test the hypothesis that TGA1-like genes have an ancestral role in spikelet development by assessing the pattern of TGA1 protein expression in the early stages of inflorescence development across representative grass species and a closely related grass outgroup. In addition, to evaluate the role of TGA1-like gene evolution in glume thickening and rachis invagination, we generated a phylogeny of TGA1-like genes, and compared patterns of amino acid sequence variation and protein expression with inferred shifts in inflorescence morphology. Finally, to distinguish between alternative hypotheses for tga1 diversification between teosinte and maize, we compared mRNA and protein expression patterns in early- and late-stage inflorescences of both subspecies.
Materials and Methods
Plant material and growth conditions
Seeds of maize W22 and Mo17, Avena strigosa (Preston 18), sorghum (Sorghum bicolor) (Malcomber 3116), Eleusine indica (PI217609), green millet (Setaria viridis) (PI204624) and Tripsacum dactyloides (Kellogg 4) were obtained from the United States Department of Agriculture (USDA), or collected in the field (T. dactyloides), and grown at the University of Missouri – St. Louis or the University of Wisconsin – Madison (USA) (maize Mo17) at 20–22°C under constant light. Seeds of Streptochaeta angustifolia (Malcomber 3123) were collected by Lynn G. Clark and grown at the University of Missouri – St. Louis on vermiculite in a humid chamber under constant light. Inflorescence material from teosinte (PI384065, PI384071, Ames21814) was harvested from plants grown at the University of Missouri – Columbia by Sherry Flint-Garcia. Other species included in the phylogenetic and expression analyses were grown from seed stocks at the University of Missouri – St. Louis under standard glasshouse conditions.
Sequencing and phylogenetic analysis
Partial TGA1-like, ndhF, phyB and waxy sequences were isolated from genomic DNA of representative Andropogoneae and non-Andropogoneae panicoids, nonpanicoid grasses and the nongrass outgroup Joinvillea ascendens (Joinvilleaceae) listed in Supporting Information Tables S1 and S2. Total DNA extractions were performed following Malcomber & Kellogg (2006). ndhF, phyB and waxy genes were amplified as described previously (Mathews et al., 2002). TGA1-like genes were amplified using the forward primers 482F (5′-AGTGCAGCAGGTTCCATCTACT-3′), 681F (5′-GATSAAAACCGAGGAGAGYCC-3′) or 1000F (5′-GACTCSGAYTGTGCTCTCTCTC-3′) and the reverse primer 1368R (5′-TACTGCCAYGAGAASGGC-3′). Each primer combination amplified exons two and three, part of exon two and exon three, and part of exon three, respectively. Cycling parameters were 94°C for 5 min, followed by 30 cycles of 94°C for 30 s, 55°C for 30 s and 72°C for 1 min, with a final extension of 72°C for 10 min. PCR fragments were purified and cloned into the pGEM-T easy vector (Promega, Madison, WI, USA). Plasmid DNA was cleaned by alkaline lysis and sequenced using the plasmid primers T7 and SP6. Sequencing reactions were carried out using the BigDye 3.1 terminator cycle sequencing protocol (Applied Biosystems, Foster City, CA, USA), and analyzed on an ABI capillary sequencer (Applied Biosystems). Sequence quality was assessed using phred (Ewing et al., 1998).
Sequences with phred scores above 20 were manually aligned in MacClade 4 (Maddison & Maddison, 2003). Phylogenetic analyses were performed on the TGA1 nucleotide alignment and a combined alignment of ndhF, waxy, phyB, tb1 and TGA1-like genes using GARLI 0.951 and MrBayes 3.1.2 following model optimization in MrModelTest (Ronquist & Huelsenbeck, 2003; Nylander, 2004; Zwickl, 2006). Maximum likelihood (ML) analyses were run using 10 random addition sequences with 500 bootstrap replicates, implementing the HKY + I + Γ (TGA1) or GTR + I + Γ (combined) model of evolution based on the results from MrModelTest 2.2. Bayesian analyses were run twice for ten million generations, sampling every 1000 generations, with 25% of trees discarded as burn-in. Gaps were treated as missing data. Trees were rooted with orthologous sequences from the nongrass outgroup J. ascendens (TGA1-like) or Pennisetum alopecuroides (Paniceae) and Axonopus fissifolius (Paniceae) (combined) and supplemented with previously generated sequences from GenBank (Tables S1, S2).
To determine the identity of residue six (the amino acid that varies between teosinte and maize tga1), a 460-bp fragment was amplified from maize, teosinte, Bouteloua gracilis (K-1976-198) (Chloridoideae) and Rottboellia aurita (Panicoideae) using the primers Tga1-position1F (5′-ATGGATTGGGATCTCAA-3′) and Tga1-SBP-Rev (5′-TTGGAGTGCGSCTCGCACACCTTGT-3′). PCR conditions were as described above. To identify potential SPL protein consensus binding sites, putative promoter and coding regions were obtained from rice (Ehrhartoideae), Brachypodium distachyon (Pooideae), sorghum (Panicoideae), Setaria italica (Panicoideae) and maize (Panicoideae) genomic sequences corresponding to the duplicated grass FUL-like genes (Fig. S1). These sequences were used to search for the core consensus sequence CCGTAC (Birkenbihl et al., 2005; Liang et al., 2008).
Character state reconstructions
Glume hardness and the extent of rachis internode invagination were determined for species available in the herbarium collection of the Missouri Botanical Gardens. Glumes were scored as membranous to coriaceous (0) or hard (1), based on the ease of penetration with a needle, whereas rachis internodes were scored as flat to terete (0) or concave (1), based on their overall shape and invagination. In order to reconstruct character state transitions, the likelihood of each glume and internode state was estimated under the Mk1 (single rate of change between states) model of evolution for each node of the best ML tree by tracing characters over the 312 most likely tree topologies in Mesquite version 2.01 (Maddison & Maddison, 2006, 2007).
Inflorescences at different stages of development were fixed in FAA (47.5% (v/v) ethanol, 5% (v/v) acetic acid, 3.7% (v/v) formaldehyde (Sigma, St. Louis, MO, USA)) using vacuum infiltration. To increase the definition of the cell walls, tissue was stained with 1% eosin Y in 95% ethanol, and dehydrated into paraffin wax, following Jackson (1991). Ribbons of 8-μm longitudinal sections were cut, mounted on Probe-On-Plus microscope slides (Fisher Scientific, Pittsburg, PA, USA) and left to dry at 37°C overnight.
In situ hybridization was performed as described by Jackson et al. (1994) using a 500-bp mRNA probe from the 3′ end of maize tga1. As the sequence of this probe was 96% identical to a closely related maize paralog (see Results), hybridization shows expression of both paralogous genes. Sense and antisense riboprobes were generated using T7 and SP6 Megascript in vitro transcription kits (Ambion Inc., Austin, TX, USA) with digoxygenin-labeled UTP (Roche, Indianapolis, IN, USA), according to the manufacturer’s instructions. Probe hydrolysis followed Jackson (1991). Probe hybridization, washing, immunolocalization and photography followed Jackson et al. (1994) and Malcomber & Kellogg (2004). Immunolocalization was performed following Lucas et al. (1995) using the TGA1 antibody described in Wang et al. (2005). Photographs were imported into Adobe Photoshop and adjusted for contrast, brightness and color balance.
Allele-specific expression assay
We employed an allele-specific assay of not1 and tga1 accumulation in the F1 hybrids of inbred maize and teosinte lines. Seven teosinte inbreds (TIL1, TIL3, TIL5, TIL9, TIL11, TIL14 and TIL25) were used as pollen parents to two or more maize inbreds (B73, CML103, Ki3, Mo17, Oh43 and W22), creating a total of 27 F1 hybrids (Tables S3, S4). Total cellular RNA was isolated from 5–10 immature ears from one or two plants of each cross; 5-μg aliquots of each of the RNA samples were DNase treated and reverse transcribed (RT) using a polyT primer and Superscript III reverse transcriptase (Invitrogen, Carlsbad, CA, USA), following the manufacturer’s instructions. The integrity of the cDNA was checked using 0.5 μl of the RT reactions as the template for PCRs with the actin primers WH182 (5′-CCAAGGCCAACAGAGAGAAA-3′) and WH183 (5′-CCAAACGGAGAATAGCATGAG-3′). The same actin primers were also used to check for genomic DNA contamination. In all cases, the actin primer check was negative for genomic DNA contamination. The RT reactions were diluted and 1-μl aliquots were employed as the template for PCRs using two sets of fluorescently labeled primers: TNW120F (5′ FAM labeled; 5′-ATCCTGCCCCGCCGTGCAG-3′) and TNW121R (5′-CACGAGAAGGGCATCGACGACGAG-3′) to amplify not1, and TNW124F (5′ FAM labeled; 5′-GCGATTCTCACCATTTGCGCATC-3′) and TNW126R (5′-AGGCGTGGCGGCTCCCAG-3′) to amplify tga1 (TAQ Core Kit, Qiagen, Valencia, CA, USA). PCR products were assayed on an ABI 3700 fragment analyzer and peak areas were determined using Gene Marker version 1.70.
In order to perform the allele-specific expression assay, we utilized the fact that the maize and teosinte parents have different allele sizes for both not1 and tga1. Both pairs of TNW primers flank an indel in the respective gene and amplify different sized products for the maize and teosinte parents. The relative message level associated with the two chromosomes in each sample was calculated as the ratio of the area under teosinte/maize allele peaks. The same assay was also performed with the DNA from each of the plants used for RNA extraction to assess any bias in allele amplification in PCR. This analysis revealed that the ratio of the maize/teosinte allele was on average 1.63 for not1 and 1.10 for tga1, indicating that the teosinte allele was slightly under-represented in both cases. Measurements for the area of the teosinte alleles were therefore multiplied by the factor determined by DNA from the particular plant, and the final message level ratios were calculated with the corrected teosinte areas.
Phylogenetic analysis and amino acid diversity of TGA1-like genes
Homologs of tga1 were amplified and sequenced from 26 taxa, comprising 25 grasses and one nongrass outgroup (J. ascendens, Joinvilleaceae) (Table S1). In genes of all species, the binding site for microRNA156 (Corngrass1 (Cg1)) (Chuck et al., 2007) was fully conserved. ML and Bayesian phylogenetic analyses were largely concordant. Well-supported relationships of TGA1-like genes outside the Andropogoneae (ML bootstrap above 70%, posterior probabilities above 95%) largely followed known generic relationships (GPWG, 2001; Sánchez-Ken et al., 2007) (Fig. 1), except that Orthoclada laxa (Panicoideae) was sister to PACMAD grasses, although without support; we interpret this as an artifact created by sparse sampling outside Panicoideae. Within Andropogoneae there was little resolution, and internal branches were relatively short (Fig. 1). However, the predicted sister relationship between the genes of maize and teosinte was strongly supported (100% ML bootstrap; 100% posterior probability), and there was moderate to strong support for an Andropogoneae clade excluding Coix lacryma-jobi and S. bicolor (70% ML bootstrap; 100% posterior probability) (Fig. 1).
In contrast with all other sampled taxa, maize and teosinte tga1 had a duplicate locus that was found to be located on the same chromosome, c. 270 kbp from tga1. Given the proximity of this tga1 duplicate, we have named it neighbor of tga1 (not1). tga1 (designated GRMZM2G10511) and not1 (designated AC233751.1_FG002) are located at positions 44 508 235–44 512 898 and 44 779 972–44 784 721, respectively, on chromosome 4 of the maize B73 Reference Genome version 2. Sequencing from the maize inbred line W22 revealed 87% amino acid identity between tga1 and not1, and these fell into two well-supported sister clades, each containing a sequence from maize and teosinte (Fig. 1). Thus, the two Zea TGA1-like genes are inferred to have arisen from a duplication event after the divergence of Zea and Tripsacum, but before the divergence of teosinte from modern maize.
A section of TGA1-like genes spanning the start codon and part of the highly conserved SBP domain was amplified from maize (not1), teosinte (not1), B. gracilis and R. aurita, and aligned with previously sequenced genes from O. sativa, S. italica, S. bicolor, teosinte (tga1) and maize (tga1). This region contains the lysine to asparagine change (position six) between tga1 orthologs of maize and teosinte, and the amino acid (position five) that was mutated in maize by ethyl methanesulfonate mutagenesis, resulting in a teosinte-like phenotype (Wang et al., 2005). In all cases, residue five was a phenylalanine, as in both maize and teosinte tga1, and residue six was a lysine, as in teosinte tga1, indicating that this region is highly conserved across Andropogoneae and grasses as a whole.
To determine whether TGA1-like proteins have the potential to regulate spikelet development through the direct regulation of genes known to be involved in spikelet development, FUL-like genes from multiple grasses were examined for the SPL protein consensus binding sequence CCGTAC (Birkenbihl et al., 2005; Liang et al., 2008). For every target species, one or more consensus binding sites were found within the putative promoter and/or intronic region of each FUL1 and FUL2 gene (Fig. S1). For example, in the rice FUL1 gene (OsMADS14) sensuPreston & Kellogg (2006), three binding sites (italic type) were found, one (TTCCGTACGA) 4410 bp upstream of the protein-coding start site and two (CCCCGTACGA and CTCCGTACTT) within the first intron; in the rice FUL2 gene (OsMADS15), three (GACCGTACGA, GCCCGTACCA and GCCCGTACCA) binding sites were found 394, 3965 and 4028 bp upstream, respectively, of the protein-coding start site. By contrast, a putative SPL binding site was only found for one of the FUL3 genes examined. The rice FUL3 gene (OsMADS18) had one binding site (CACCGTACCC) 5232 bp upstream of the protein-coding start site; no consensus binding sites were found within the putative promoter and intronic regions of S. italica, S. bicolor or Z. mays FUL3 orthologs.
Andropogoneae phylogeny and inflorescence trait evolution
In order to estimate the number and direction of shifts in glume hardening and rachis internode invagination within Andropogoneae, we supplemented previously generated ndhF, waxy, phyB and tb1 sequences with newly generated TGA1 sequences, and inferred species relationships (Spangler et al., 1999; Lukens & Doebley, 2001; Mathews et al., 2002) (Table S2; Fig. S2). We found no strongly supported topological differences between trees on the basis of individual markers, and therefore carried out phylogenetic analyses based on combined sequences (Spangler et al., 1999; Lukens & Doebley, 2001; Mathews et al., 2002; this study). The best ML tree based on combined analyses resolved some of the internal relationships within Andropogoneae with high support, including a close relationship between Coelorachis afraurita, Coelorachis selloana, Rottboellia species and Phacelurus huillensis, and between Hemarthria species and Heteropholis sulcata (Fig. S2). Similar to previous analyses based on individual markers, deep branches of the tree were very short and showed little support between groups (Spangler et al., 1999; Lukens & Doebley, 2001; Mathews et al., 2002). For example, the position of Elionurus muticus was unstable within the larger Coelorachis/Zea clade (Fig. S2; data not shown). This supports the hypothesis of a rapid radiation during the diversification of Andropogoneae subtribes (Celarier, 1956; Spangler et al., 1999).
We used the best ML tree based on the combined dataset to reconstruct the evolution of glume and inflorescence axis traits within Andropogoneae based on ML (Fig. 2). In the case of glume morphology, membranous to coriaceous glumes are inferred as the ancestral state of the Andropogoneae. However, hard glumes are inferred to have evolved between four to seven times independently, followed by four to five independent secondary reductions in glume hardness (Fig. 2). Similar to maize, the secondary reduction of glume hardness in P. huillensis, Urelytrum digitatum and the Hemarthria/Heteropholis clade was strongly supported by both morphological and phylogenetic data; it remains unclear whether coriaceous glumes were secondarily derived in E. muticus and Chionachne koenigii.
Similar analyses for inflorescence axis morphology reconstructed the ancestor of Andropogoneae as having flat to terete internodes, with four to six transitions to invaginated internodes across the phylogeny (Fig. 2). Morphological data clearly support the independent loss of rachis internode invagination in maize and E. muticus. However, topological uncertainty makes it equally likely that flat internodes are ancestral or, instead, evolved secondarily in E. muticus.
TGA1-like protein expression
In order to test whether TGA1-like proteins have an ancestral role in spikelet/flower development, and whether changes in their expression correlate with changes in glume hardening and invagination of the inflorescence axis, we compared expression patterns across representative grasses and a closely related outgroup (J. ascendens) that vary in inflorescence morphology (Figs 3, 4). Expression patterns were consistent with an ancestral role for TGA1-like genes in spikelet development, but did not support the hypothesis that modification of TGA1-like protein expression underlies shifts in glume and rachis internode morphology.
Joinvillea ascendens has conventional monocot flowers comprising outer tepals, inner tepals, stamens and a gynoecium (Preston et al., 2009). Expression of JaTGA1 was evident in all organs of the J. ascendens flower, but was absent from the subtending floral bract and branch (Fig. 3a). Unlike J. ascendens and other grasses, the early diverging species S. angustifolia does not have tepals and lacks true spikelets and glumes. However, previous studies have inferred bracts 1–5 to be transformationally homologous to glumes, and bracts 6–8 to be transformationally homologous to outer tepals and lemmas/paleas (Preston et al., 2009). In early S. angustifolia flower development, SaTGA1 was detected in bracts 1–11, but was absent from the inflorescence branch base, presumed to be homologous to a bract subtending the floral unit (Preston et al., 2009) (Fig. 3b). In mid- to late stages of inflorescence development, SaTGA1 was no longer detectable in bracts 1–6, but was strongly expressed in bracts 7–11 and reproductive structures (Fig. 3c).
In grasses with true spikelets, TGA1-like protein was detected in both outer and inner glumes, and within all floret organs at early stages of development, regardless of glume and inflorescence axis morphology (Fig. 3). In the pedicellate spikelets of A. strigosa, AsTGA1 was expressed in the membranous glumes, and in all organs of the bisexual proximal florets and reduced distal florets (Fig. 3d). Likewise, in mid-stage development of E. indica and S. viridis spikelets, EiTGA1 (Fig. 3e) and SvTGA1 (Fig. 3f) were expressed in all floral organs. However, at this stage, very little protein was detectable in the membranous subtending glumes of both species.
Within Andropogoneae, all sampled species, with the exception of S. bicolor, are monoecious. In S. bicolor, SbTGA1 was expressed in all organs of both the bisexual sessile spikelets that have very hard glumes, and the male/sterile pedicellate spikelets that have coriaceous glumes, at early to mid-stages of spikelet development (Fig. 3h,i). In C. lacryma-jobi, ClTGA1 was expressed in the coriaceous glumes and floral organs of male and female C. lacryma-jobi inflorescences (Fig. 3g). Similarly, TdTGA1 protein was detectable in hard glumes and all floral organs of young T. dactyloides female spikelets (Fig. 3j). Finally, cumulative TGA1/NOT1 expression was evident in young to mid-stage glumes, inflorescences axes and all floral organs of both female and male maize (Fig. 3k,l) and teosinte (Fig. 3m,n) spikelets.
In contrast with early- to mid-stage development, late-stage patterns of protein expression differed markedly among T. dactyloides, teosinte and maize (Fig. 4). In late-stage spikelet development of T. dactyloides, little TdTGA1 protein was detected in the glumes of male (Fig. 4a) or female (Fig. 4b) spikelets, or within maturing floral organs of female spikelets (Fig. 4b); expression was still detectable in late-stage male flowers in the lodicules and stamens (Fig. 4a). Furthermore, at both early and late stages of development, TdTGA1 protein expression was barely detectable in the rachis (Figs 3j, 4a,b). In silk-stage ears of maize, TGA1/NOT1 protein was expressed within the wall of the invaginated rachis, but was only weakly detectable in glumes, and was undetectable in gynoecia (Fig. 4f,g). By contrast, although only expressed at low levels in gynoecia, lodicules, paleas and lemmas, TGA1/NOT1 in silk-stage teosinte ears was strongly expressed in the small cells of the abaxial mesophyll of glumes and the rachis (Fig. 4c–e). This pattern of teosinte TGA1/NOT1 expression was maintained following fruit set. No staining was detectable in control sections lacking the TGA1/NOT1 antibody (Fig. 4h) and the results were consistent between three independent experiments.
tga1/not1 mRNA expression in Zea
To determine whether tga1/not1 expression is distinct from the expression of the corresponding proteins, a single antisense mRNA probe was used to assess the cumulative expression pattern of tga1/not1 mRNA in maize and teosinte. The expression of tga1/not1 was similar for maize and teosinte ears at early stages of development (Fig. 5a,d). Before floral differentiation, tga1/not1 mRNA was detected in the inner and outer glumes, and within the spikelet and floret meristems. At later stages, tga1/not1 was expressed throughout all floral organ primordia, but was excluded from the expanded region at the base of the outer glume (Fig. 5a,d).
At the silk stage, coincident with thickening of the glumes and rachis invagination in teosinte, tga1/not1 expression was markedly different between maize and teosinte (Fig. 5b,e,f). In maize, tga1/not1 was barely detectable in all well-developed spikelet organs and in the invaginated rachis (Fig. 5b). However, in teosinte, tga1/not1 was abundantly expressed in the large parenchymal cells of the adaxial glume mesophyll and the rachis, but was undetectable in the small cells of the abaxial glume mesophyll (Fig. 5e,f).
In order to differentiate between elevated tga1 vs not1 expression in teosinte relative to maize ears, an allele-specific expression assay was carried out (Tables S3, S4). For tga1, the average corrected ratio of the teosinte/maize allele expression was 1.06 with a standard deviation of 0.21 (Table S3). Thus, the expression of tga1 was not significantly different between teosinte and maize (P =0.122). By contrast, for not1, the average corrected ratio of the teosinte/maize allele expression was 2.02 with a standard deviation of 1.23 (Table S4). Although the F1 values varied substantially in allele-specific expression, on average, teosinte not1 expression was significantly higher than maize not1 expression (P =3.51 × 10−5).
Several species of Andropogoneae develop hard indigestible structures that surround the growing fruit, potentially allowing them to escape high levels of seed predation and to optimize seed dispersal (Wilkes, 1967). In some species, protection is afforded by a hardened (indurate) lower glume that wraps around the developing fruit (e.g. Sorghum), whereas, in other species, the fruit is shielded by both an indurate glume and a strongly invaginated (sunken) rachis internode (e.g. Tripsacum, Coelorachis and teosinte) (Clayton & Renvoize, 1986; E. A. Kellogg, pers. obs.). Together with previous studies, our morphological and phylogenetic analyses suggest that indurate glumes have evolved one to three times in the clade containing Tripsacum and Heteropholis, with at least four secondary transitions back to coriaceous glumes (Urelytrum, P. huillensis, maize and Hemarthria sp.) (Spangler et al., 1999; Lukens & Doebley, 2001; Mathews et al., 2002; Watson & Dallwitz, 1992; Teerawatananon et al., 2011; this study). Similar character reconstructions for invaginated rachis internodes are less apparent. However, they clearly support at least two independent origins of flat to terete internodes, once at the base of Arundinella plus Andropogoneae and once at the base of maize, and demonstrate that internode invagination and glume hardness can be uncoupled (Fig. 2).
Evolution and function of TGA1-like genes across grasses
Morphological and phylogenetic evidence suggests that hardened glumes and invaginated inflorescence axes have been gained and lost more than once within the Andropogoneae. However, it is unknown whether mutations in orthologous genes are responsible for the independent losses of these traits. In the case of maize, mutations in tga1 are tightly associated with the evolution of glumes that are membranous, at least apically, and flat inflorescence axes (Dorweiler et al., 1993; Wang et al., 2005). Furthermore, various lines of evidence suggest that the causative mutation underlying the evolution of these traits is within the coding region of tga1 (Wang et al., 2005). Analysis of the TGA1-like gene from multiple species that vary in glume and rachis morphology has revealed a conserved lysine at amino acid position six, as previously found in the TGA1-like gene of rice and representative species of Zea (except maize) (Wang et al., 2005). Similarly, the microRNA binding site is fully conserved in all TGA1-like genes sampled, suggesting purifying selection and conservation of the negative interaction between miR156 (Cg1) and TGA1 (Chuck et al., 2007). Together, these findings suggest that variation in TGA1-like amino acid sequences is not correlated with independent losses of hard glumes and invaginated internodes within Andropogoneae. However, as we obtained only partial sequences for many grass species, we cannot rule out the possibility that changes in TGA1-like genes have been important for morphological evolution in these species.
An alternative mechanism to explain the reduction in glume hardness and invaginated internodes is evolution at the level of TGA1-like gene regulation. This hypothesis predicts differences in protein expression between species that vary in glume and rachis internode morphology. Analyses of protein expression across grasses outside of Zea that have membranous to coriaceous glumes (S. angustifolia, A. strigosa, E. indica, S. viridis, C. lacryma-jobi, pedicellate S. bicolor and male T. dactyloides), indurate glumes (sessile S. bicolor and female T. dactyloides), flat or terete rachis internodes (S. angustifolia, A. strigosa, E. indica, S. viridis and C. lacryma-jobi) and invaginated rachis internodes (T. dactyloides) revealed no consistent pattern between morphology and protein expression. Except for maize, this is consistent with functional conservation for TGA1-like proteins across grasses.
The fact that a single amino acid change can simultaneously affect multiple traits suggests that tga1 is an upstream regulator of the Z. mays inflorescence developmental pathway (Wang et al., 2005), and that the development of Z. mays glumes and inflorescence axes is tightly coupled. However, the exact function of grass TGA-like genes is unclear. Our data show that TGA1-like genes have a conserved expression pattern in grass spikelets, being expressed early in spikelet and floral meristems, and at mid-stages of development in glumes and floret organ primordia. Furthermore, in the grass outgroup J. ascendens, which has more typical monocot flowers, JaTGA1 is present in all developing floral organs, but is not detectable in the subtending floral bract. Together, these protein expression patterns are similar to the combined expression patterns of FUL-like genes, MADS-box transcription factors related to the floral meristem and floral organ identity genes APETALA1 (AP1) and FUL of Arabidopsis thaliana (Mandel et al., 1992; Mandel & Yanofsky, 1995; Ferrándiz et al., 2000; Preston & Kellogg, 2007; Preston et al., 2009). Indeed, in A. thaliana and Antirrhinum majus, tga1 (SPL) homologs directly regulate FUL-like genes in leaves and shoot apical meristems (Klein et al., 1996; Wang et al., 2009; Yamaguchi et al., 2009; Preston & Hileman, 2010). Because SPL protein binding site sequences were found in all examined FUL1 and FUL2 genes from across the grass family, we posit that the regulatory interaction between SPL and FUL-like genes is conserved between core eudicots and grasses. As rice has 19 SPL proteins, most or all of which are expressed in inflorescences (Xie et al., 2006; Yang et al., 2008), future analyses of FUL-like gene expression in tga1-like silenced lines will be required to specifically test the hypothesis of TGA1–FUL interaction.
TGA1 and domestication of the maize ear
Probably the most striking example of glume and rachis evolution is between the ears of maize and its ancestor teosinte. Although teosinte has indurate glumes and strongly invaginated inflorescence axes, reduced hardness and flat axes have been selected for in maize glumes. Previous studies have revealed an important role for tga1 in the domestication of maize (Dorweiler et al., 1993; Wang et al., 2005). These morphological differences were associated with six fixed differences in the promoter and one amino acid difference in the protein-coding region of the SPL gene tga1 (Wang et al., 2005).
Several lines of evidence suggest that the amino acid difference in tga1 underlies the inflorescence differences between maize and teosinte. First, the lysine of teosinte in this position has been conserved through grass evolution, including taxa as disparate as rice (Ehrhartoideae), wheat (Pooideae) and R. aurita (Panicoideae) (Wang et al., 2005; this study). Indeed, extensive sequencing from multiple teosinte populations found this to be the only fixed difference between maize and teosinte; the six promoter polymorphisms are variable within teosinte (Zhao, 2006). Second, a single nonconservative mutation in an amino acid adjacent to the potentially causative amino acid of maize tga1 results in teosinte-like glume and rachis internode structures (Wang et al., 2005). Finally, the identification of a tga1 paralog (not1) explains the apparent differences between the tga1 gene and TGA1 protein expression patterns in maize and teosinte reported in Wang et al. (2005). Specifically, quantitative (q)RT-PCR analyses previously showed no difference in tga1 RNA levels between early- to silk-stage maize and teosinte ears (Wang et al., 2005). By contrast, we found a quantitative difference in RNA levels between maize and teosinte lower glumes and rachis internodes by in situ hybridization (Wang et al., 2005). Protein levels were also different between the two, as measured by western blots and immunolocalization (Wang et al., 2005). The qRT-PCR primers are specific to tga1 alone, and therefore we conclude that gene expression is indeed similar between maize and teosinte. However, the in situ probe and antibody probably capture both tga1/TGA1 and not1/NOT1 expression, so that elevated RNA and protein levels in late-stage teosinte ears reflect higher expression of not1/NOT1 rather than tga1/TGA1. This interpretation is strongly supported by our allele-specific expression data. We infer that the expression of the tga1 gene and its protein product are thus quantitatively similar between the two species. Therefore, our data do not support the hypothesis that differences in the teosinte and maize glume and rachis are a result of variation in protein stability or translational efficiency. Instead, we prefer the hypothesis that the replacement of either residue five or six of TGA1 changes its biochemical function and causes the glume and rachis differences between maize and teosinte.
We thank Sherry Flint-Garcia for providing flowering material of teosinte, and Shelby Kleweis, Jimena Nores and Chris Gillespie for help with sequencing. This research was funded by the National Science Foundation grants DBI-0820619 (J.D.) and DBI-0110189 (E.A.K.).