Out of 20 progeny of a mutagenic plant B2-8A-2-18 (F₂ generation) derived from a cross between iAc and DsG transgenic lines 31, we found a mutant plant (B2-8A-2-18-16) with panicle characteristics similar to a previously reported EMS-induced frizzy panicle mutant fzp 20 and a γ-ray induced frizzy panicle mutant fzp2 15. We have designated this mutant branched floretless 1 (bfl1).

The panicle of a bfl1 mutant looks like a stick and its rachis-branches never stretch because of tangling (Fig. 1B). Development of the primary rachis-branches appeared to be normal (Fig. 1C). The panicle length of wild-type and the bfl1 mutant were 15.5 cm and 14.9 cm, respectively and the number of primary rachis-branches of wild-type and the mutant were 6.1 and 6.4, respectively. However, lateral and terminal spikelets of primary and secondary order rachis-branches failed to produce florets. Instead, they continued to produce next-order rachis branches in alternate axis (Fig. 1G, 2E) most likely after initiation of bract-like structures (Fig. 1F, 1G, 2E, 2F). As observed in the scanning electron mirgrographs (SEMs), the bract-like structure in the mutant (Fig. 2F,2G,2H) corresponds to rudimentary glumes of wild-type (Fig. 2B,2C,2D) because they have comparable positions and surface appearances. The mutant showed no difference from the wild-type plant with respect to vegetative growth, or to the time of transition from vegetative phase to reproductive phase.

PCR analyses showed that the original bfl1mutant contained both iAc and Ds (Table 2) and was negative for GFP (the T-DNA or Ds launching pad selection marker), indicating that the transposed Ds and iAc have not segregated but the original Ds launching pad has segregated away. As iAc was present, excision of Ds and possible reversion to wild-type phenotype could be expected. The mutant plant was therefore propagated vegetatively to recover seeds produced after reversion events.

In such vegetatively-propagated plants, several revertant sectors were found within a tiller, or in different tillers (Fig. 3A, 3B). In total, 15 normal mature seeds (Fig. 3E), 11 sterile and abnormal seeds (Fig. 3F, 3G) and 6 incomplete spikelets (Fig. 3H) were obtained. Out of these 15 putative revertant seeds, 10 germinated (PR1 to PR10) and were grown to maturity. PR2, PR5 and PR7 showed normal panicle development and partial sterility (Fig. 3C; Table 2), whereas the other 7 plants (PR1, PR3, PR4, PR6, PR8, PR9 and PR10) still displayed the bfl1 phenotype (Fig. 3D; Table 2). Ds-specific PCR analysis showed positive results for PR2 to PR10 and a negative result for PR1 (Table 2). The latter was attributed to a possible Ds3' end truncation as indicated in the subsequent Southern blot hybridization analyses (see below).

Ds3' (LW1125; GenBank Acc. No. message URL http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=nucleotide&list_uids=33771666&dopt=GenBank&term=bfl1AY343496) and Ds5' (LW1455; GenBank Acc. No. message URL http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=nucleotide&list_uids=33771665&dopt=GenBank&term=bfl1AY343495) flanking sequences were rescued from the bfl1 mutant and PR1 using TAIL-PCR 19 and plasmid rescue system, respectively. Initial searches (February 2002) failed to identify any homologous sequences from GenBank (NCBI) database but the China Rice Genome database http://btn.genomics.org.cn/rice/ contained one entry (contig 239) showing homology to bfl1 flanking sequences. PCR characterization of the bfl1 locus and gene prediction were therefore based on contig 239. Subsequently, the Ds insertion in the bfl1 mutant was mapped to rice chromosome 7 corresponding to PAC clone P0625E02 (GenBank Acc No. message URL http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=nucleotide&list_uids=34395191&dopt=GenBank&term=ap004570AP004570).

Southern blot hybridization analyses with the gus probe (which hybridizes with the 3' end of the Ds insertion) showed that the bfl1 mutant had a single-copy of Ds showing one 7.2 kb positively-hybridizing band (Fig. 4). PR3, PR4, PR6, PR8, PR9 and PR10 also had the same 7.2 kb band and showed mutant panicles, indicating that they were homozygous Ds insertion plants resulting from self-pollination of gametes with Ds in the bfl1 locus. PR1, however, produced mutant panicles but did not show any positively-hybridizing band, suggesting a deletion of the 3' region of Ds comprising the gus gene sequences. PR2 and PR5 did not have the same 7.2 kb band, but had a band of a different size indicating that they were homozygous revertants with Ds excised from the bfl1 locus and re-inserted into other locations in the rice genome. PR7 had one extra positively-hybridizing band besides the 7.2 kb band, indicating that it was a heterozygous revertant resulting from fertilization of a gamete with Ds in the bfl1 locus by a gamete with a re-transposed Ds. Thus, normal panicle-development appeared to result from excision of Ds from the bfl1 locus. Based on the sequences of contig 239 and the PAC clone P0625E02, the expected size of the positively-hybridizing band in the Southern blot analyses of HindIII-digested bfl1 mutant plant DNA with thegus probe is 9.3 kb. However, the observed band size was ~7.2 kb. A close observation of the published sequence of this fragment revealed that a single nucleotide substitution could result in an additional HindIII restriction site, and a shorter positively-hybridizing band. We tried locating the possible nucleotide substitution in various sources of genomic DNA including a freshly amplified DNA but failed to see any nt transition. Most likely this particular transition varies between different samples of Nipponbare. The position of the HindIII RE recognition site formed due to possible transition is indicated in Fig. 6. Nevertheless, our genomic sequences flanking both Ds3 (GenBank Acc. No. message URL http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=nucleotide&list_uids=33771666&dopt=GenBank&term=bfl1AY343496) and Ds5 (GeneBank Acc. No. message URL http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=nucleotide&list_uids=33771665&dopt=GenBank&term=bfl1AY343495) ends unambigously maps the location of the Ds insertion.

Three sets of PCRs were performed to confirm the presence or absence of a Ds insertion in the bfl1 locus: PCR1 to amplify the region flanking the Ds insertion; PCR2 and PCR3 to amplify part of the 5' and 3' ends of the Ds element and their flanking genomic regions, respectively (Table 1; Fig. 6). PR2 and PR5 were positive for PCR1 and negative for PCR2 and PCR3 confirming that they were homozygous revertants. PR7 was positive for PCR1, PCR2 and PCR3 confirming that it was a heterozygous revertant (Table 2). The genotypes of PR2, PR5 and PR7 were confirmed by subsequent analyses of progeny plants. The PCR results of PR1 confirmed that it had a Ds3' truncation (Table 2). All other plants were negative for PCR1 and positive for PCR2 and PCR3, suggesting that there had been no excision of the Ds element from the bfl1 locus.

Taken together, the above analyses strongly suggest that the insertion of the Ds is the cause of the bfl1 mutation and the excision of the Ds from the bfl1 locus is associated with the reversion of the mutation.

It is a well known fact that during Ds insertion it creates 8 bp host sequence duplication. However, when the Ds is re-excised it often leaves footprints of this direct repeat sequences. Ds excision footprints of PR2, PR5, PR7 and PR12 showed retention of 4, 7, 0 and 7 bp of the duplicated host sequence, respectively (Fig. 5). Although the proper reading frame could be restored with the excision of Ds in PR7, the reading frame would not be restored for PR2, PR5 and PR12, suggesting that the Ds insertion was not in the coding region of the BFL1 gene, but in its non-coding or regulatory region.

Subsequently, we obtained additional revertants, including a whole normal panicle with 34 seeds. When 26 seeds of this panicle were germinated and grown to maturity, they showed segregation of homozygous un-excised Ds (Ds/Ds), heterozygous Ds excision (Ds/+) and homozygous Ds excision (+/+) in the ratio of 1:2:1 (6:13:7). All Ds excision plants had the same excision footprint as PR12 (Table 2 and Fig. 5), indicating that there had been germinal excision of Ds during the transition from vegetative to reproductive stage.

Analyses of 10-kb sequence of the ~30-kb contig 239 from the China Rice Genome database harboring the BFL1 locus by gene prediction programs FGENESH http://www.softberry.com/berry.phtml/ and GENSCAN http://genes.mit.edu/GENSCAN.html identified a single-exon gene capable of encoding a protein with the DNA binding domain of the EREBP/AP2 family of plant transcription factors 2636, 1515 bp downstream from the Ds insertion. Subsequently, the same gene was predicted in the rice PAC clone P0625E02 sequences (Fig. 6). The predicted message URL http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=nucleotide&list_uids=34899319&dopt=GenBank&term=P0625E02.25P0625E02.25 protein corresponds to our predicted protein of 318 amino acids. In view of the function of EREBP/AP2 domain genes, and the excision of Ds from the bfl1 locus resulting in revertants with normally-developed panicles, we concluded that this EREBP/AP2-domain gene is likely to be the BFL1 gene. Furthermore, the loss or reduction of the BFL1 expression in mutant plants was confirmed by gene-specific RT-PCR using total RNA isolated from leaf and developing panicles (Fig. 7). BFL1 was expressed in wild-type leaves and developing panicles but was greatly reduced in the bfl1 mutant. The expression of BFL1 in revertant panicles was evident in the RT-PCR analysis (Fig. 7). However, we could not backup the RT-PCR result with Northern blot analysis possibly due to very low expression level.

The EREBP/AP2 domain of BFL1 was found to be identical to the ERF domain of a recently-isolated maize BD1 gene and that of a rice BD1-like gene 9. The EREBP/AP2 domain of BFL1 also shared a high level of similarity with those of class II ERFs, such as LEAFY PETIOLE and TINY of Arabidopsis 3237 and, as in LEAFY PETIOLE, the EREBP/AP2 domain of BFL1 was located close to the N-terminus of the protein (Fig. 8).

In rice, immediately after the transition to the reproductive stage, the SAM switches into an IM, from which primary rachis-branch meristems (PBMs) initiate in a spiral phyllotaxy. The IM degenerates after a given number of PBMs have been initiated. Concurrently, primary rachis-branches differentiate into secondary rachis-branch meristems (SBMs) or lateral spikelet meristems (LSMs) alternatively, and are terminated by terminal spikelet meristems (TSMs). SMs produce two pairs of glumes, rudimentary and empty glumes, as well as a single floret composed of lemma, palea, lodicules, stamens and carpel 29. Therefore, the rice inflorescence or panicle represents a determinate branched structure composed of a main spike bearing primary and secondary rachis-branches, lateral and terminal spikelets (Fig. 1A). Each rice spikelet contains a single floret, a pair of empty glumes and a pair of rudimentary glumes in contrast to the pair of glumes and two florets present in maize but it is still to be resolved whether empty glumes or rudimentary glumes are analogous to maize glumes 51329.

In the bfl1 mutant, although the development of the SAM is normal until primary rachis-branches (PBs) are produced, lateral and terminal spikelets are replaced by secondary rachis-branches (SBs) which then continually initiate higher order branches in a distichous phyllotaxy resulting in an indeterminate branched panicle. Based on our observation, the rudimentary glumes of spikelets seemed to be initiated normally as bract-like structures in the bfl1 mutant (Fig. 1G, 2F), but subsequent SMs ceased development and failed to produce empty glumes and FM, instead, their identity was converted to that of a branch meristem (BM), indicating that SM identity is acquired but fails to transit from SM to FM. Therefore, we propose that the function of BFL1 is to repress BM identity within the spikelet, or to positively regulate the initiation of FM. Similarly, in the maize bd1 mutant, ear SMs could not transit to FMs and continue to produce glumes and spikelet-like structure 10. As we believe that the spikelet initiation has occurred in bfl1 it is logical to assume that the bract-like structure (rudimentary glume) are in fact actual glumes as discussed previously 29.

Public availability of the near-complete rice genome sequences (http://www.ncbi.nlm.nih.gov/;http://rgp.dna.affrc.go.jp/; http://btn.genomics.org.cn/rice; http://portal.tmri.org/rice/) allowed us to locate the single Ds insertion in the bfl1 mutant on rice chromosome 7 corresponding to the PAC clone P0625E02 and indicated that the gene most likely to have been affected by this Ds insertion is one which encodes an EREBP/AP2 domain-like protein (message URL http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=nucleotide&list_uids=34899319&dopt=GenBank&term=P0625E02.25P0625E02.25). Since this domain is identical to the ERF domains of the recently reported maize BD1 gene and the rice BD1 – like gene 9 we conclude that BFL1 is the rice ortholog of maize BD1.

Although ERF proteins are known as plant-specific transcription factors involved in the ethylene response 24, they are also implicated in plant development 43237. The direct evidence for the role of ERF proteins in organogenesis comes from the characterization of the maize BD1. The BD1 gene mediates the transition from a spikelet to a floret meristem during maize ear development 910 and its expression is temporally and spatially regulated during spikelet and floret initiation 9. Strong phenotypic resemblance between the maize bd1 mutant and our rice bfl1 mutant further supports our claim that BFL1 is the rice ortholog of BD1.

Failure to detect BFL1 transcript by Northern blot analysis was most probably due to its low expression. It is not uncommon to have such a low expression levels of transcription factor genes such as BFL1. However, expression studies using RT-PCR under optimal conditions indicated substantial reduction in the BFL1 transcript levels in the bfl1 mutant compared to wild-type and was restored in the revertant (Fig. 7). Further studies on spatial and developmental expression patterns of BFL1 in wild-type, mutant, and revertant plants are needed to define the regulation of BFL1 expression.

Based on our prediction, the transcription of BFL1 starts at 121 nt upstream of the translation start site (TSS). A TATA box and the Ds insertion are located 157 and 1515 nt upstream of the TSS, respectively, suggesting that the Ds insertion is not in the core promoter region of BFL1 gene. The observed mutant phenotype and the RT-PCR data, however, suggest that the Ds insertion severely affects the expression of the BFL1 gene, thus suggesting that the region upstream of the core promoter is required for efficient transcription initiation of the BFL1 gene. The Ds insertion in bfl1 is most likely to block assembly of the transcription initiation complex, which includes not only the core promoter but also proximal and distal enhancer elements. Further investigation into such regulatory elements is required to unravel the transcriptional control of BFL1. It is obvious that there are several other genes involved in this complex regulatory pathway and identification of these genes is crucial to understand the three-step transition unique to grasses i.e. branch to spikelet to florets.

All rice lines used in this study were derived from the japonica cultivar Nipponbare. The bfl1 mutant described in this paper was recognised initially among F₃ generation plants derived from a cross between the iAc (pSK300, TT3-24-1-1) and Ds gene trap (DsG, pSK200, TT2-10-1-1) transgenic lines described previously by Upadhyaya et al. (2002). The iAc and DsG binary vector constructs, tissue culture and Agrobacterium transformation procedures used have been described previously 3031. Plants were grown under controlled glasshouse conditions with 25 ± 3°C day and 21°C night temperatures with 16 h day length.

Genomic DNA was extracted from plants using the PureGene nucleic acid isolation kit (Gentra Systems Inc. Minneapolis, MN, USA) according to the manufacturer's instructions. Ds5' and Ds3' flanking sequences were rescued using the built-in plasmid rescue system 31 and TAIL-PCR 19, respectively. For plasmid rescue, ~2 μg of genomic DNA was digested with SacI, then extracted with phenol/chloroform, precipitated and self-ligated in 500 μl of ligation mix containing 5 Weiss units of T4 DNA ligase at 16°C overnight. The ligated DNA was used for electroporation after ethanol precipitation. Plasmid clones were analysed by appropriate restriction enzyme analyses before being selected for sequencing. TAIL-PCR was performed using three nested primers Ds3_1, Ds3_3, Ds3_6587⁺ and an arbitrary degenerate primer AD2 (Table 1) according to Liu and others 19 with minor modifications. The purified PCR product was re-amplified using Ds3_6587⁺ and AD2 to ensure success of the sequencing reaction. Sequencing was performed with the reagents of the ABI Prism BigDye termination cycle sequencing kit (Applied Biosystems, Foster City, CA, USA) according to the manufacturer's instructions.

DNA sequence comparisons were performed using the BLAST program 12 searching against the NCBI GenBank http://www.ncbi.nlm.nih.gov/ and China's indica rice sequence database http://btn.genomics.org.cn/rice. The ORF was identified by using FGENESH http://www.softberry.com/berry.phtml/ and GENSCAN http://genes.mit.edu/GENSCAN.html. Alignment of EREBP/AP2 domains was performed using programs of Genetics Computer Group Wisconsin software suit 11.

For Southern blot hybridization analysis, genomic DNA (~10 μg) was digested with HindIII, fractionated on a 0.7% agarose gel, and blotted onto a Hybond-N⁺ membrane (Amersham Life Science, England) according to the manufacture's instructions. The gus gene of the DsG construct was used to prepare radioactively-labelled probes with the Megaprime^TM DNA labelling system (Amersham Life Science, England) according to the manufacturer's instructions. The membrane was hybridized at 42°C for 6 h, washed at 60°C with 0.1 × SSC and 0.1% SDS and visualized by autoradiography using a phosphor imager (Molecular Dynamics, Sunnyvale, CA, USA).

PCR primers (Table 1) were designed using the 'prime' program of the Genetics Computer Group (GCG) Wisconsin software suit 11. Ds insertion plants were analysed by Ac- (primers Ac_1931⁺ and Ac_2382^-) and Ds- (primers GUS_313^- and GPAInt) specific PCR amplification. The following PCR program was used: 30 cycles of 94°C for 2 minutes, 58°C for 30 seconds, and 72°C for 1 minute, followed by a final 72°C for 5 minutes. To identify the presence or absence of Ds insertion in the bfl1 locus, a set of primers annealing to the flanking region of the Ds insertion were designed based on the published rice genomic sequences (contig 239 of China Genome database at http://btn.genomics.org.cn/rice. Three sets of PCRs were then performed: PCR1 with primers LW1125_For and LW1125_Rev amplified the region flanking the Ds insertion. PCR2 with primers LW1125_Rev and Ds5_112^-and PCR3 with LW1125_For and Ds3_6587 ⁺amplified part of the 5' and 3' ends of the Ds element and their flanking genomic regions, respectively (Table 1; Fig. 6). The PCR conditions were 30 cycles of 94°C for 2 minutes, 60°C for 30 seconds, and 70°C for 1 minute, followed by a final 70°C for 5 minutes.

BFL1 transcript levels in RNA samples of mutant and wild-type were visualized by RT-PCR with BFL1 specific primers (Table 1) using OneStep RT-PCR kit (QIAGEN Inc, California, USA) reagents according to manufacture's instructions. Primers specific to the rice sucrose synthase gene RSs1 35 was used as an internal control for RNA integrity and DNA contamination.