Orthologs of Arabidopsis thaliana stomatal bHLH genes and regulation of stomatal development in grasses

Plants conquered land approximately 400 million years ago(Edwards et al., 1998). Correlated with this expansion in habitat was the development of an epidermis that, although made highly impermeable by a lipid-rich cuticle, still permitted the exchange of external CO₂ for internal O₂and water vapor. Microscopic epidermal valves called stomata were the structural innovations that allow this regulated exchange(Edwards et al., 1998). Stomata are present on the aerial surfaces of all large land plants. At a minimum,stomata consist of two guard cells, a pore and an underlying airspace. In many species, however, the stomatal complex includes subsidiary cells adjacent to the guard cells. Both guard and subsidiary cells are morphologically distinct from other epidermal cell types. Since their appearance in the fossil record(contemporaneous with the appearance of land plants), stomatal densities and distributions have changed significantly, but guard cell morphology has remained quite constant (Edwards et al.,1998). In general, there are only two broad classes of stomatal guard cells: the kidney-shaped cells found most plant species and the dumbbell-shaped guard cells found in grasses(Evert, 2006) (see Fig. 1A).

Grass stomata, as described as early as 1881(Campbell, 1881), have both a pair of dumbbell-shaped guard cells and associated subsidiary cells. Grass stomata are usually arranged in linear files and this final arrangement reflects the developmental process that created them(Fig. 1B). Grass stomata may be on either the top (adaxial) or bottom (abaxial) side of the expanded leaf, but in many species they are preferentially or exclusively abaxial.

Grass guard cell morphology is thought to be derived from the kidney-shaped guard cells found in mosses, ferns, gymnosperms, dicots, and some monocots. For convenience, we will refer to this form as `dicot'. Dicot stomata often lack subsidiary cells, but they may also have two or more such cells. In contrast to the linear arrangement typical of grass stomata, dicot stomata are scattered on leaf surfaces, a pattern that reflects their `dispersed' mode of development (Fig. 1B). This distribution pattern is not random, however, and stomatal architecture and patterns are valuable taxonomic characters for both living and fossilized plants (Garland, 1984; Stebbins, 1960).

In Arabidopsis, stomatal development requires a series of asymmetric and symmetric cell divisions in a specialized epidermal cell lineage. Stomatal development is initiated by an asymmetric division in a protodermal cell to produce a small meristemoid and a larger sister cell. The meristemoid is a self-renewing cell and can continue asymmetric divisions. However, it possesses only transient `stem cell-like' properties and after one to three divisions differentiates into a guard mother cell (GMC). The GMC undergoes a single symmetric division to produce a pair of guard cells(Fig. 1B) (reviewed by Bergmann and Sack, 2007). Stomatal development proceeds roughly in an apical-basal gradient with the more mature stages near the tip, but this is not absolute because sister cells of the stomatal precursors might divide later, intercalating new stomata into areas where stomata previously formed (Fig. 1B). Stomata are formed via similar developmental mechanisms on both the abaxial and adaxial leaf surfaces.

Three of the positive regulators that direct this three-step sequential stomatal development in Arabidopsis are the closely related basic helix-loop-helix (bHLH) domain transcription factors FAMA, MUTE and SPEECHLESS(SPCH) (MacAlister et al.,2007; Ohashi-Ito and Bergmann,2006; Pillitteri et al.,2007). SPCH is expressed in many young epidermal cells and controls the first asymmetric division of protodermal cells to initiate the stomatal lineage (MacAlister et al.,2007; Pillitteri et al.,2007). MUTE is highly expressed in meristemoids and is required for termination of meristemoid stem cell identity and the transition to GMC fate (Pillitteri et al.,2007). Finally, FAMA is expressed in GMCs and regulates the last stage of stomatal development by promoting the symmetric differentiation of a GMC into a guard cell pair (Ohashi-Ito and Bergmann, 2006). FAMA, MUTE and SPCH therefore act as molecular switches controlling major cell fate transitions during stomatal development.

Stomatal development in grasses can be divided into five stages(Stebbins, 1960)(Fig. 1B). Here, stomatal development exhibits a strong spatiotemporal gradient with early stages taking place in the proximal portions of the leaf and guard cells differentiating later in distal regions. In stage one, cell files that are capable of forming stomata are determined (blue shading at leaf base, Fig. 1B). Asymmetric division of cells in these files (stage two, middle leaf section, Fig. 1B) generates GMCs as the smaller daughters. A second asymmetric division then occurs in the cells adjacent to the newly specified GMCs to produce a pair of subsidiary mother cells (SMCs), so at this third stage the guard cell complex consists of a GMC and two subsidiary cells. In the fourth stage, the GMC divides symmetrically into two box-shaped guard cells. During the final stage, guard cells undergo extensive elongation and morphogenetic changes to form the final pair of dumbbell-shaped cells with a central pore between them (red cells at tip of leaf, Fig. 1B)(Sack, 1994).

Despite the differences in stomatal ontogeny, morphology and pattern between monocots and dicots, protein sequences of the key regulatory genes SPCH, MUTE and FAMA are highly conserved between representatives of these two angiosperm divisions. In this study, we identify likely orthologs of Arabidopsis SPCH, MUTE and FAMA in two grass species: rice (Oryza sativa) and maize (Zea mays). Through mutation, transgenics, and by monitoring gene expression in situ, we demonstrate that there is significant conservation of function of the FAMA gene between monocots and dicots. By contrast, although MUTE and the two SPCH genes maintain some common functions in grasses, they have diverged in their roles and domains of expression.

Arabidopsis thaliana Columbia ecotype seeds were sterilized and vernalized at 4°C for 3 days before sowing. Plants were grown at 22°C with a 16 hours light/8 hours dark photoperiod. Rice plants were Oryza sativa spp japonica cv Nipponbare (X.-W. Deng, Yale University) or mutants from RIKEN (NF7789) or POSTECH (PFG3A-52237.R). After de-husking,surface sterilization and imbibition, rice seeds were germinated on Murashige and Skoog (MS) media in magenta boxes for 12 days before transfer to soil and growth at 22°C with a 16 hours light/8 hours dark photoperiod. Maize plants were Zea mays GNN5 (V. Walbot, Stanford University); plants were germinated directly in soil in conditions of 16 hours light/8 hours dark.

DNA and amino acid sequences for rice and maize FAMA, MUTE and SPCH1/2 were retrieved by BLAST searches against the GRAMENE database(www.gramene.org/multi/blastview)and TIGR database(http://tigrblast.tigr.org). Protein sequences of FAMA, MUTE and SPCH orthologs were aligned using CLUSTALW(version 7.0.9) software, with manual adjustment using Bioedit to align the bHLH motif and the C terminus. Intron/exon structure and splicing patterns of rice and maize bHLH genes were obtained from the GRAMENE database. Additionally, we searched each protein manually for the presence of additional predicted functional domains using multiple software tools available from PROSITE(http://ca.expasy.org/prosite/). A phylogenetic tree was constructed with the aligned stomatal bHLH protein sequences using MEGA [version 3.0; http://www.megasoftware.net/index.html(Kumar et al., 1994)] with a 1000 bootstrap replicates.

Total RNA was isolated from washed whole 10-dpg rice and maize seedlings using the RNeasy Kit (Qiagen, Valencia, CA, USA). cDNAs were generated by SuperScript III reverse transcriptase (Invitrogen, Carlsbad, CA, USA). The 2.72 kb genomic region including coding sequences for OsFAMA was amplified from rice genomic DNA. OsMUTE (684 bp) and ZmMUTE(618 bp) were amplified from rice and maize cDNA, respectively (for primers,see Table 1). A partial clone of OsSPCH1 (AK287482) was obtained from the Rice Genome Resource Center(http://cdna01.dna.affrc.go.jp/cDNA/). OsSPCH2 cDNA was produced by RACE PCR amplification performed according to the manufacturer's (Invitrogen) instructions using OsSP2-1F and OsSP2-1R for the 5′ end, and OsSP2-2F and OsSP2-2R for the 3′ end(for primers, see Table 1). Products from amplifications were cloned in pENTR/D-TOPO (Invitrogen) and sequenced. Overexpression constructs were built by recombining each pENTR clone into the binary vector pH35SGS (Kubo et al., 2005). All constructs were introduced into Arabidopsis Columbia ecotype by Agrobacterium tumefaciens-mediated transformation(Clough and Bent, 1998). Transgenic T1 plants were selected on agar-solidified MS-containing hygromycin. When possible, T2 lines were analyzed for phenotypes, but severe phenotypes (arrest and sterility) generated by the expression of some transgenes sometimes necessitated the use of T1s.

Hybridizations were performed as described(Long and Barton, 1998) with minor modifications. Digoxigenin-labeled RNA probes were directed against the unique N-terminal region of OsFAMA (1064 bp), the complete cDNA of OsMUTE and ZmMUTE, the unique third exon (621bp) of OsSPCH1 and the N terminus (489 bp) of OsSPCH2 (for primers,see Table 1). Sense RNA probes were used as negative controls. Stomatal lineage-expressed OsSCR1(Kamiya et al., 2003) was used as a positive control.

Pro_FAMA::OsFAMA was constructed by combining the 2.72-kb OsFAMA genomic DNA fragment with the 2.5-kb fragment upstream of FAMA gene (Ohashi-Ito and Bergmann, 2006) in pMDC107(Curtis and Grossniklaus,2003). The construct was transformed into a fama-1/+ line by agrobacterium-mediated transformation. Similar strategies to construct Arabidopsis regulatory regions driving rice or maize cDNAs were used to test complementation of mute and spch mutants. For each case, T1 transformant lines were selected for hygromycin resistance. T2s from these plants that segregated mutant phenotypes were selected and examined microscopically. The number of phenotypically wild-type versus mutant seedlings (seedlings devoid of stomata) was counted. These results were analyzed using the χ² test from R statistical analysis software. A segregation ratio of 0.25 indicated failure to rescue (reduced to 0.0625 if the construct rescued). All rescue data reported were significant(P≤0.05) and at least 12 T2 lines were tested for each construct.

To test expression levels of 35S::OsMUTE and 35S::MUTE,total RNA was extracted using Trizol (Invitrogen) and approximately 500 ng was used in cDNA synthesis reactions using Superscript III Reverse Transcriptase(Invitrogen; for primers, see Table 1).

Scanning electron microscopy and DIC images were obtained using protocols reported by Ohashi-Ito and Bergmann(Ohashi-Ito and Bergmann,2006), except that confocal images were acquired using a Leica TCS SP5 confocal microscope. Images were analyzed using ImageJ software and prepared for publication using Adobe Photoshop and Illustrator.

Sequence data from this article can be found in the TAIR and GRAMENE databases under accession numbers At3g24140 (FAMA), At3g06120(MUTE), At5g53210 (SPCH), At3g26744 (SCRM),At1g12860 (SCRM2), Os05g50900 (OsFAMA), Os05g51820(OsMUTE), Os06g33450 (OsSPCH1), Os02g15760(OsSPCH2), Os11g03110 (OsSCR1), Os11g32100(OsSCRM1), Os01g71310 (OsSCRM2), AZM5_84542(ZmFAMA), AZM5_28767 (ZmMUTE), AZM5_15260 (ZmSPCH1)and AZM5_13800 (ZmSPCH2).

Protein sequences of SPCH, MUTE and FAMA are highly similar in the bHLH(putative DNA binding) domain and in an essential C-terminal region(MacAlister et al., 2007) that has previously been shown to mediate dimerization in other bHLHs(Feller et al., 2006) (see Tables S1 and S2 in the supplementary material). Despite strong similarity within these two domains, SPCH, MUTE and FAMA each contain unique features,allowing orthologs to be assigned unambiguously. FAMA and SPCH contain N-terminal acidic domains (with limited sequence similarity between them, see Fig. S1 in the supplementary material). This domain in FAMA has been demonstrated to activate transcription(Ohashi-Ito and Bergmann,2006). SPCH contains an ∼90 amino acid domain that is phosphorylated by mitogen activated protein kinases (MAPKs) and is referred to as the MAPK target domain (MPKTD) (Lampard et al., 2008). MUTE encodes neither of these domains (see Fig. S2 in the supplementary material).

To identify homologs of SPCH, MUTE and FAMA within the angiosperms, we searched numerous databases for genomic and EST sequences,concentrating especially on the grasses (GRAMENE; http://www.gramene.org/). Putative orthologs of SPCH, MUTE and FAMA were designated based on protein sequence identity using the entire protein and by comparing intron/exon structures and synteny when possible. Fig. 1C shows a tree of SPCH, MUTE and FAMA sequences from Arabidopsis,maize and rice. FAMA and MUTE are present as single copy genes in these species, whereas two SPCH genes are found in each of the grasses (Fig. 1C). The unique features of SPCH, MUTE and FAMA, notably the long N terminus of FAMA and the MPKTD of SPCH are readily apparent in the grass homologs (see Fig. S1 in the supplementary material). Inspection of the genomic microstructures surrounding FAMA and MUTE indicates a high degree of microsynteny between Arabidopsis and rice(Fig. 1D); genes adjacent to MUTE and FAMA are homologs of the genes adjacent to OsMUTE and OsFAMA. For example, At3g06140, a ring-finger gene, and At3g06270, a protein phosphatase 2C (PP2C), are adjacent to MUTE. In rice, OsMUTE is flanked by the ring-finger gene Os05g51780 and the PP2C gene Os05g51510(Fig. 1D).

The maize and rice genomes each encode a single gene homologous to FAMA. OsFAMA (Os05g50900) encodes a 493 amino acid protein (see Fig. S1 in the supplementary material). Within the bHLH domain, OsFAMA is 94%(47/50 amino acids) identical to FAMA. The maize ortholog ZmFAMA (AZM5_84542)is 92% identical to Arabidopsis, and the identity between OsFAMA and ZmFAMA is 92% (46/50 amino acids; see Tables S1 and S2 in the supplementary material). The grass FAMA genes also share with Arabidopsis FAMA a single intron within the bHLH domain at a conserved site (not shown). Although there is limited amino acid identity among the N-terminal extensions of the FAMA genes, these regions might serve similar functional roles as both rice and maize FAMA have glutamine- and proline-rich stretches that are typical of transcriptional activation domains (see Fig. S1 in the supplementary material).

If rice and Arabidopsis FAMA are required for the same developmental transition, then we would expect OsFAMA to be expressed in the leaf epidermis during the GMC to GC transition stages. Furthermore,loss-of-function mutations in rice fama should result in the absence of mature guard cells, while maintaining the normal overall pattern of precursors. RNA in situ hybridization was performed to determine the tissue localization of OsFAMA. In wild-type seedlings, OsFAMAtranscript was first detected in the leaf epidermis and vasculature of the sheath elongation zone (SEZ, Fig. 2B,C). In the leaf blade expansion zone (BEZ, Fig. 2B), OsFAMAtranscript was most intense in the abaxial epidermis of the sheath(Fig. 2D,E). OsFAMAwas not detected in the shoot apical meristem (SAM) or in leaf primodia at the earliest stages of leaf development (sheath division zone, SDZ, Fig. 2B,F). This expression pattern is consistent with a gene whose role is to regulate the latest stages of stomatal development.

We identified a transposon insertion mutant in OsFAMA (NF7789, Rice Genome Project of NIAS). The osfama-1 mutant allele harbors an insertion in the third exon (Fig. 3A). We confirmed the presence of the insertion by PCR using primers flanking the insertion site (for primers, see Table 1) and, in homozygous mutants, we did not detect OsFAMAtranscript (Fig. 2G-I),indicating that osfama-1 is probably a loss-of-function allele. The homozygous osfama-1 mutants displayed a severe dwarf phenotype and yellowish leaf blade. At 10 days post-germination (dpg), osfama-1homozygotes were only ∼30% of the height of their wild-type siblings(Fig. 3B). After transplanting to soil, osfama-1 plants continued to be smaller than their wild-type or heterozygous siblings, and died before reaching reproductive maturity(Fig. 3C).

The epidermis of the osfama-1 plants showed an intriguing defect in stomatal development. The linear arrays of stomatal complexes in osfama-1 mutants were indistinguishable from those of wild type in terms of numbers and overall pattern (Fig. 3D, compare with 3F). However, instead of their typical dumbbell morphology, the two guard cells were box-shaped(Fig. 3E, compare with 3G). Each mature osfama-1 guard cell resembled a GMC or a very immature wild-type guard cell. We interpret this to mean that OsFAMA is required for guard cell differentiation. In contrast to the phenotype of Arabidopsis fama-1 mutants, however, the rice GMCs did not undergo extra cell divisions.

Because both osfama-1 and Arabidopsis fama mutants have defects in GC production, but Arabidopsis and rice GCs have different morphologies, we tested whether OsFAMA could promote the same outcomes as FAMA when expressed in Arabidopsis. Overexpression of FAMA induces the formation of excess, unpaired guard cells in the epidermis of Arabidopsis(Ohashi-Ito and Bergmann,2006). We generated transgenic Arabidopsis expressing OsFAMA under the control of the cauliflower mosaic virus 35S promoter(35S::OsFAMA). Like overexpression of FAMA(Fig. 3H), overexpression of OsFAMA could induce the production of unpaired guard cells(Fig. 3I), but did so less effectively than FAMA. We then expressed OsFAMA under the control of the FAMA promoter (Pro_FAMA::OsFAMA)and transformed this construct into heterozygous fama-1 mutant plants. Among the fama-1 homozygous mutant progeny from twelve independent T2 transgenic lines, we were able to identify plants with mature guard cells (Fig. 4C). These data indicate that FAMA functions to promote guard cell differentiation in representatives of both the monocots and the dicots despite the morphological divergences of stomatal guard cells in these plant groups.

Stages in stomatal development controlled by SPCH and MUTE in Arabidopsis are not easily compared with stages in grass stomatal development. Nonetheless, homologs of these earlier acting bHLHs are readily found. MUTE is a small protein, lacking both an N-terminal extension and a MPKTD (see Fig. S2 in the supplementary material). Homologs from maize and rice are ∼80% identical in the bHLH domain and are ∼50%identical over the whole protein (see Tables S1 and S2 in the supplementary material). Interestingly, although MUTE is not a predicted or actual substrate of the MAPKs that phosphorylate SPCH(Lampard et al., 2008), OsMUTE and ZmMUTE posses multiple potential MAP kinase phosphorylation sites (see Fig. S1 in the supplementary material). In contrast to the single SPCH of the dicots Arabidopsis, Poplar and Ricinus,SPCH is represented by two genes in rice and maize(Fig. 1C). One of the two rice SPCH genes (OsSPCH2) is much more similar to SPCHthan the other (see Tables S1 and S2 in the supplementary material), but all the grass SPCH genes are larger than SPCH, and the 483 amino acid peptide encoded by OsSPCH1 is markedly so (see Figs S1 and S2 in the supplementary material). Most of the increased length in OsSPCH1 is in a 190 amino acid N-terminal extension that has little similarity to other SPCH proteins. OsSPCH2 and both maize SPCH paralogs have smaller N-terminal domains containing a conserved transcriptional activation domain. The unique MPKTD region of SPCH (relative to MUTE and FAMA) is easily identified in the grass SPCH homologs and the sites demonstrated to be phosphorylated in SPCH are conserved (see Fig. S1 in the supplementary material) (Lampard et al.,2008).

We were unable to detect OsSPCH1 or OsSPCH2 transcripts in rice tissues from the seedling stage through leaf maturity by in situ hybridization, RT-PCR or RNA blot analysis (not shown). This is not surprising given that SPCH is found at low levels at a transient stage in leaf development (MacAlister et al.,2007). We queried the whole-genome transcriptional profiles of rice cell types isolated by laser-capture microdissection (RICEATLAS, http://bioinformatics.med.yale.edu/riceatlas/search.jspx). OsSPCH transcript was detected only in the coleoptile (a covering over embryonic leaves). By contrast, OsMUTE and OsFAMA were readily detected in multiple young leaf sample tissues, consistent with the in situ data.

Although we cannot conclude from expression data whether either SPCH homolog is involved in rice stomatal development, we do have recourse to several functional assays - characterization of an insertional mutant in rice,and cross species complementation and overexpression studies. Overexpression of SPCH leads to ectopic divisions in pavement cells(MacAlister et al., 2007). No phenotypes were observed with 35S::OsSPCH1 expression, but 35S::OsSPCH2 was able to promote cell division in the pavement cells(see Fig. S3 in the supplementary material). We next tested the ability of OsSPCH2 (the paralog with activity in the overexpression assay and with higher overall similarity to SPCH) to rescue spch;however, Pro_SPCH::OsSPCH2 plants failed (0/16 lines) to rescue the spch mutant phenotype.

POSTECH insertion mutant lines of osspch1 (PFG_1D-04011.R) and osspch2 (PFG_3A-52237.R) were obtained, and we could confirm the presence of an insertion in the osspch2-1 line(Fig. 5A; for primers, see Table 1). Plants homozygous for the osspch2-1 insertion were green, but slightly smaller in size than their wild-type siblings (data not shown). Microscopic examination of the leaves revealed an appreciable decrease in the number of stomata(Fig. 5C,D) and the presence of stomatal patterning abnormalities (Fig. 5F). This phenotype is found only in the plants homozygous for the insertion (2/15). Taken together, the functional assays suggest that OsSPCH2 plays a role in promoting the early events of rice stomatal development, but is not completely equivalent to SPCH.

By in situ hybridization, OsMUTE mRNA first accumulates as rice leaf primordia initiate (Fig. 6B,C). In sections nearest to the base of the leaf, strong expression was observed in the whole P1 primordium, and weak expression was occasionally detected in the P4 leaf (Fig. 6D). In sections that include P1 primordia that enclose the SAM, OsMUTE remains strongly expressed in the P1 primordia; in addition,strong expression was also found in the SAM proper and in P2 primordia(Fig. 6E). Transcript accumulates much more strongly in stomatal precursor cells as the P2 leaves mature, whereas the P1 leaves maintain strong expression in all epidermal cells (Fig. 6G). When compared with OsFAMA (e.g. Fig. 2B), it is clear that OsMUTE is expressed at earlier developmental stages.

OsMUTE expression in rice leaf primordia is spatially and temporally distinct from that in Arabidopsis. To test whether this reflected differences in the cis-regulatory elements of the MUTEpromoters or differences in other factors (different plant morphologies or trans-acting regulatory factors), we made a Pro_OsMUTE::GUSreporter (for primers, see Table 1). In 20 independent T2 lines we observed Pro_OsMUTE::GUS activity broadly at young stages in the developing Arabidopsis leaf primordia (see Fig. S4B in the supplementary material). GUS staining was also observed in the stomatal lineage cells, but it was not restricted to meristemoids (see Fig. S4C in the supplementary material). This expression pattern differs from that of Pro_MUTE::GUS, whose expression is observed first at cotyledon tips (Pillitteri et al.,2008), and is then restricted to meristemoids(Pillitteri et al., 2007). The Pro_OsMUTE::GUS expression in leaves resembles the broad early expression pattern of Pro_SPCH::GUS(MacAlister et al., 2007).

We tested whether OsMUTE could complement mute by expressing OsMUTE cDNA with the endogenous MUTE promoter or the 35S promoter. OsMUTE expressed under either promoter could partially rescue mute, as evidenced by altered segregation ratios in the self progeny (see Materials and methods) and by the presence of guard cells at the edges of leaves exhibiting the typical mute phenotype(arrested meristemoids) in the center (Fig. 4D).

In Arabidopsis, overexpression of MUTE leads to the conversion of epidermal cells to GMCs that then divide to form paired guard cells (MacAlister et al.,2007; Pillitteri et al.,2007). We overexpressed OsMUTE (35S::OsMUTE) in Arabidopsis and compared overexpression phenotypes with those produced by 35S::MUTE. In both cases a spectrum of phenotypes was observed (Fig. 7); the phenotypical disruptions correlated with the level of overexpression, as assayed by RT-PCR (Fig. 7A,I;OL1, OL2 and OL3 of 35S::OsMUTE transgenic lines and AL1, AL2 and AL3 of 35S::MUTE plants). At the highest expression levels (AL3 and OL3 lines), both MUTE and OsMUTE converted all epidermal cells in the cotyledons and leaves into guard cells(Fig. 7H,L). These lines suggest that OsMUTE is capable of inducing the same phenotypes as MUTE when expressed at very high levels.

The epidermis of lines expressing lower levels of OsMUTE looked qualitatively different than lines expressing MUTE at comparable levels (Fig. 7A,I). By quantifying the type and number of cells produced by either construct, we found that both could increase total epidermal cell number above that of wild type, but there was a significant difference in the types of cells produced in each case (see Fig. S5 in the supplementary material). Moderately expressed 35S::MUTE (AL2, Fig. 7C) induced the formation of large guard cell clusters(Fig. 7K) among a few normal jigsaw-shaped pavement cells. By contrast, moderately expressed 35S::OsMUTE (OL2) led to the production of small, non-stomatal marker-expressing cells at the expense of pavement cells(Fig. 7O). We observed the same trend in epidermal phenotypes in lines with low expression levels of 35S::MUTE (AL1) and 35S::OsMUTE (OL1; see Fig. 7B,N,P; see also Fig. S5 in the supplementary material).

Moderate expression of OsMUTE induces cell division, but the resultant cells do not all become guard cells, a phenotype that is strikingly similar to the phenotypes created by the expression of SPCH variants with altered MPKTDs(Lampard et al., 2008). We examined this connection between OsMUTE and SPCH by expressing OsMUTE in an spch-3 (null) background. This enabled us to assay the rescue of spch by OsMUTE and to carefully characterize the effects of OsMUTE expression in an epidermis lacking stomatal lineage cells(Fig. 8A). 35S::MUTE;spch-3 plants (four independent T2 lines) produced pavement cells and clusters of mature stomata (Fig. 8B) (Lampard et al.,2008). By contrast, 35S::OsMUTE; spch-3 homozygotes from five independent T2 lines exhibited a common phenotype of ectopic epidermal cell divisions. In mature (12-dpg) cotyledons from these lines, the typical crenulated pavement cells were replaced by small cells, but no stomata were produced (Fig. 8C).

Because OsMUTE and MUTE did not behave in the same way,and because no insertion mutations in OsMUTE could be verified, we expanded our investigation into the behavior of grass MUTE genes by analyzing MUTE from maize. By RNA in situ hybridization, ZmMUTE was expressed in the SAM and in emerging leaf primordia (Fig. 9C-G), with the highest transcript levels in the leaf margins of P1-P4 (Fig. 9G). This expression pattern - strong in immature, mitotically active leaf zones - is similar to that of OsMUTE. However, at later developmental stages, ZmMUTE RNA expression is broader than that of OsMUTE(Fig. 9E, compare with Fig. 6F). Like OsMUTE,the ZmMUTE cDNA driven by the endogenous MUTE promoter(Pro_MUTE::ZmMUTE) can complement Arabidopsis mute-1 (12 T2 lines, see Materials and methods).

In Arabidopsis, 35S::ZmMUTE expression resulted in both stomatal and whole plant phenotypes (Fig. 10A-D). 35S::ZmMUTE was exceedingly effective at promoting stomatal formation (Fig. 10B-E). The effects on stomatal production were similar to those caused by the strongest lines of 35S::MUTE(MacAlister et al., 2007; Pillitteri et al., 2007);however, the effect on seedling morphology was unique. Because 35S::ZmMUTE phenotypes were seedling lethal, we also generated estrogen-inducible lines (Pro_Est::ZmMUTE). Seedlings germinated on MS plates and transferred at 12-dpg to media containing 5 μM estrogen for 12 hours showed excess cell proliferation (visible after 24 hours, Fig. 10F) and stomatal production (after 48 hours, Fig. 10G; 72 hours, Fig. 10H). It appears that ZmMUTE behaves much like MUTE in its ability to convert many cell types (including non-stomatal lineage cells) into guard cells.

Our data provide the first functional evidence that homologs of stomatal genes identified in Arabidopsis can regulate stomatal formation in the grasses. OsFAMA is closely related to the Arabidopsis FAMA gene in sequence and expression pattern, and complements the Arabidopsis mutant. Both fama-1 and osfama-1mutants fail to complete the final stage of stomatal formation. These findings suggest that the bHLH transcription factor FAMA acts at a high level in a transcriptional regulatory cascade - in essence specifying the transition to guard cell fate, not the details of guard cell morphology. This role is consistent with previous reports that ectopic expression of FAMAcould override previous differentiation status and induce guard cell characteristics in root epidermal and leaf internal mesophyll cells(Ohashi-Ito and Bergmann,2006).

Although there is conservation between the functions of Arabidopsis and rice FAMA, there are differences in the arrest phenotype of Arabidopsis and rice fama mutants. Most notably, osfama-1 mutants do not produce excess tumor-like cells in the stomatal cell rows. One plausible explanation for this difference is tied to the difference in stomatal lineage ontogeny. In Arabidopsis, GMCs differentiate from continuously (asymmetrically) dividing meristemoid cells and must change their division program to undergo a single symmetric division to form the guard cell pair. In grasses, GMC formation is one of the earlier stages in development and is succeeded by the recruitment of subsidiary mother cells (SMCs) from neighboring cell files. Therefore, grass GMCs are less transient cell types and are involved in several important signaling steps before they differentiate into guard cells. Rice GMCs might have intrinsically different cell cycle control features that have already acted to prevent excessive GMC division upstream of OsFAMA activity.

SPCH and MUTE control the initiation and termination of the stem cell-like meristemoids(MacAlister et al., 2007; Pillitteri et al., 2007). In the expression and functional studies on the rice and maize homologs of these genes, we found some evidence for their conserved roles in stomatal development; however, there were also some significant differences. Overexpression of OsSPCH could induce ectopic cell divisions and osspch2-1 mutants have fewer stomata than do wild type [similar to the phenotype of the weak spch-2 allele(MacAlister et al., 2007)],suggesting that SPCH has a general role in promoting early lineage cell divisions. However, OsSPCH2 was not capable of complementing spch,indicating that cell fate promoting roles might have diverged.

Both OsMUTE and ZmMUTE are expressed during stages in which the stomatal cell files are forming, suggesting that MUTE acts at an earlier stage of stomatal development in grasses than in Arabidopsis. Consistent with these findings, a GUS reporter driven by OsMUTE 5′ regulatory regions in Arabidopsis exhibited an SPCH-like expression pattern in young leaves. Although both ZmMUTE and OsMUTE can partially complement mute, their overexpression phenotypes are substantially different than that of MUTE. Expression of 35S::OsMUTE in an spch background allowed us to uncouple cell division from cell fate promotion. In plants with no stomatal lineage, MUTE can drive cells to a GMC (and later stomatal) fate; however, OsMUTE produces primarily cell divisions, a phenotype that is not only different from that of MUTE, but which in fact resembles the phenotypes produced by MPKTD-altered variants of SPCH(Lampard et al., 2008).

Divergent behavior of MUTE homologs might be expected if we consider the developmental events that precede stomatal formation in dicots and monocots. In Arabidopsis, it appears that any protodermal cell is capable of undergoing asymmetric stomatal lineage-forming divisions. Mitotic activity is not limited to a specific spatiotemporal domain and stomatal lineage cells act as dispersed `stem cell-like' populations. By contrast, monocots exhibit a very orderly and directional pattern of stomatal development and do not utilize stem cell-like divisions(Stebbins, 1960). The functions of SPCH and MUTE in Arabidopsis are tied to the initiation and termination of the stem cell-like divisions in meristemoids. In the grasses, a SPCH-like function (initiating asymmetric divisions) happens with MUTE-like timing (immediately preceding GMC formation), therefore the activity of OsMUTE could be reasonably hypothesized to be a hybrid between that of Arabidopsis SPCH and MUTE (Fig. 11).

These hypotheses are useful for considering the difference between monocots and dicots; however, we also see differences in the behavior of MUTE between the two monocots. Distinctive behaviors of rice and maize MUTE again could reflect differences in the MUTE proteins or in leaf development between the two grass species. ZmMUTE contains HBD and HTH domains that often mediate the dimerization of DNA-binding regulatory proteins(Kwok et al., 1999), and a PTB domain known to function as an adaptor to organize signaling complexes(George et al., 2008). However, no predicted protein-interaction domains are found in OsMUTE. It might be that MUTE is required to interact with partners in maize (and Arabidopsis), but not in rice. In terms of leaf development, rice leaf primordia initiate only after the prior two leaves extend, producing a shoot apex with no more than three developing leaves at a given time. Maize produces leaf primordia more rapidly, so that at least five developing leaves are always apparent at the shoot apex(Sylvester et al., 2001). If MUTE is required during specific time periods when the stomatal lineage is initiating, then the different leaf initiation rates in the two grasses might require different levels or activities of MUTE.

Understanding how SPCH, MUTE and FAMA behave also requires a consideration of their potential partners. In Arabidopsis,SCREAM (SCRM; also known as ICE1) and SCRM2,are required for multiple stages in stomatal development and might act by forming heterodimers with SPCH, MUTE and FAMA(Kanaoka et al., 2008). The presence of rice genes with significant similarity to SCRM(Os11g32100) and SCRM2 (Os01g71310) implies that modules for the transcriptional regulation of stomatal development could be conserved. Additionally, the conserved MPKTD in OsSPCH1/2 and ZmSPCH1/2indicates that post-translational regulation of the bHLHs by MAPKs might be widespread.

The details of monocot stomata developmental pathways remain to be elucidated, but can benefit from comparison to the better characterized networks in Arabidopsis. Some aspects of grass stomatal development,such as the recruitment of subsidiary cells, however, are not accessible through comparison with Arabidopsis. In the future, generating targeted knockouts of grass genes whose homologs were identified in Arabidopsis, as well as forward genetic screens in the grasses will be invaluable in defining the elements that create and pattern stomata.