Oryza sativa japonica variety Zhonghua No.11 (abbreviated as ZH11) was used as wild-type. All the plants used in this study were grown in the green house with an 8 h light and 16 h dark cycle; or in a paddy field under natural conditions in summer.

For overexpression of miR172s, genomic fragments containing the Osa_miR172s coding regions were, respectively, cloned into the pCAMBIA1301 under the 35S promoter and NOS terminator (p130135SNOS).

For overexpression of AP2s and OsMADS1, the full length cDNA of the respective genes were cloned into the p130135SNOS vector.

For overexpression of AP2-2 in the OsMADS1RNAi plants, the whole expression cassette of 35s-cDNA-nos was digested from the p130135SNOS-AP2-2 and cloned in the p230135SNOS vector, using HindIII and EcoRI.

For OsMADS1 RNAi construction, a gene-specific cDNA fragment was cloned into the p1301RNAi vector in the sense orientation using BamHI and KpnI and antisense orientation using SacI and SpeI.

Plasmid p2301-AP2-2 was transformed into OsMADS1RNAi plants with G418 selection, other plasmids were transformed into ZH11 using Agrobacterium-mediated genetic transformation with hygromycin selection, respectively (Hiei et al., 1994).

Total RNAs were extracted from the leaves or young panicles of the plants using the TRIzol (Invitrogen), and then reverse transcribed using ReverAce (TOYOBO). cDNA was synthesized from 2 μg of total RNA treated with DNase I (TOYOBO) and used as template.

Young panicles were fixed in 4% paraformaldehyde PBS solution (0.1% Triton-X-100, 0.1% Tween-20, 4% formaldehyde, 25% glutaraldehyde) overnight at 4°C, dehydrated through a concentration grade of ethanol, cleared through a dimethylbenzene series, infiltrated through a series of paraffin (Sigma-Aldrich), and finally embedded in 100% paraffin melted at 60°C. The samples were sectioned longitudinally at 7 μm and then mounted on RNase-free glass slides (Sigma). A gene-specific region of OsMADS1 was cloned into the pBSK(-) vector, linearized, and used as template for amplifying digoxigenin-labeled sense and antisense RNA probes using a DIG RNA labeling kit (Promega). In situ hybridization was performed as previously described (Coen et al., 1990).

For miRNA in situ hybridization, materials were prepared as usual. miR172 was detected with Locked Nucleic Acid (LNA) probes which were Digoxin 5′-end labeled (Exiqon).

Approximately 30 μg of total RNA was separated on 15% polyacrylamide denaturing gels. RNAs were transferred to Amersham Hybond^®-N⁺ membranes and cross-linked by UV irradiation; the membranes were hybridized with biotin-labeled DNA probes complementary to the miRNA sequences at 42°C overnight. The membranes were then washed and incubated with a stabilized streptavidin-horseradish peroxidase at 42°C. After washing with substrate equilibration buffer and adding stable peroxide solution and enhancer solution, the membranes were imaged using an FLA-5000 Phosphorimager. The DNA probes were synthesized and biotin-labeled using a 3′ end DNA labeling method.

Shoot apical meristems and IMs and young florets were decorticated under light microscope and fixed quickly in 50% FAA at 4°C overnight after vacuuming, and dehydrated through a graded concentration of ethanol. For SEM analysis, the samples were then critical point dried using liquid carbon dioxide and mounted on SEM stubs, sputter coated with gold and palladium (4:1) and examined using a SEM (Hitachi S-2460, Japan). For paraffin analysis, samples were embedded in epoxide resin and cut into slices 2–3 μm slices; strips of these slices were spread at 42°C on a hot platform overnight, stained using 0.5% toluidine Blue O and sealed for observation under the microscope (Wang J.et al., 2010).

The full length cDNA of OsMADS1 gene was cloned in frame into vector pPC86.

PCR fragments containing the binding motifs of OsMADS1 were, respectively, cloned into the p178 vector using the XhoI restriction site, which contains the CYC1 core promoter and the lacZ gene.

Yeast strain EGY48 (MATtrp1his3ura3leu2::6lexAop-LEU2) was used for transformation. The yeast assays were performed according to the manufacturer’s protocol with the substrate chlorophenol red-D-galactopyranoside (Matchmaker One-hybrid System; Clontech).

The open reading frame (ORF) of OsMADS1 was amplified and cloned into the prey vector pGAD-T7. The ORFs of AP2s were amplified and cloned into the bait vector pGBK-T7. The yeast two-hybrid assay was performed according to the manufacturer’s instructions (Clontech).

Sequence data used in this study can be found in the rice genome annotation database¹ and NCBI² under the following accession numbers: LOC_Os03g11614 (OsMADS1), LOC_Os05g03040 (AP2-1), LOC_Os03g60430 (AP2-2), LOC_Os07g13170 (AP2-3), LOC_Os06g43220 (AP2-4), LOC_Os04g55560 (AP2-5), LM379345 (Osa-miR172a), LM379346 (Osa-miR172b); LM379347 (Osa-miR172c), LM383079 (Osa-miR172d), AC091532 (actin).

Primer sequences used in this study are listed in Supplementary Table S1.

Four miR172 genes (miR172a-d) have been identified in rice genome³. To study their function, we, respectively, over expressed them under the 35S promoter through genetic transformation. For each miR172 over expression, 30–50 independent transgenic plants were gotten and over 90% of them showed the phenotypes described below. The expression levels of respective miR172s in the transgenic lines were verified by Northern blotting (Supplementary Figure S1). According to the severity of the phenotype, the transgenic plants from different miR172s overexpression could be grouped into two classes: miR172aOE, miR712cOE, and miR172dOE were similar and showed severely abnormal floral organs (Figure 1A as compared with Figure 1B), they were tentatively represented as miR172aOEs unless specified; whereas phenotypic abnormality of miR172bOE was moderate (see the following).

The miR172aOEs plants showed elongated lemma and palea, with the lemma longer than the palea (Figures 1C–E). Typically, a wild-type rice floret consists of a lemma and a palea in whorl1, and two lodicules at the lemma side in whorl2, six stamens in whorl3, and a carpel with a pair of feather-like stigmas and a green ovary in whorl4. A floret together with two pairs of sterile glumes (empty glumes and rudimentary glumes) constitutes a spikelet (Figure 1F). In the greenhouse, the lemma/palea of wild-type averaged 6.5 mm, whereas the lemma and palea of the miR172aOE was 9.07 and 7.34 mm, respectively, the lemma and palea of miR172bOE averaged 7.68 and 6.82 mm, respectively (Figure 1E).

miR172aOEs plants produced many rudimentary glumes (Figure 1C) instead of one pair in the wild-type (Figure 1F); and showed “flower-in-flower” structures (Figures 1H,I); the extreme phenotype was an overall plantlet-like structure instead of a flower (Figure 1J). All these characters indicated that miR172aOEs plants showed indeterminacy in flower development.

Phenotype of miR172bOE plants was more moderate than that of miR172aOEs. The lemma of miR172bOE was less elongated (Figures 1E,K); fewer ectopic glumes and rudimentary glumes were formed; and some flowers were fertile (Figure 1K).

Normally, the lemma and palea of wild-type plants were closed (Figure 1F) except for the short opening time during flowering. However, in the four miR172OE plants, the lemma and palea could not close due to severe distortion (Figures 1N–Q as compared with Figure 1L). As a result, the normal interlocking structure between the lemma and palea in the wild-type (Figure 1M) disappeared in the four miR172OE plants (Figures 1N–Q).

Furthermore, SEM analysis revealed that the transition from spikelet meristem to floral meristem was delayed in the miR172aOEs plants. In wild-type floret, after differentiation of a pair of rudimentary glumes and empty glumes, the floral meristem begins to development inner floral organs (Figures 2A–D). In the miR172aOEs plants, the meristem produced many rudimentary glumes before forming a floret (Figures 1C and 2E). In the process of floral meristem development, before the formation of the stamen, no visible phenotypic changes were observed (Figure 2F compared with Figure 2A, Figure 2G compared with Figure 2B), but after that, the palea/lemma elongated differently (Figure 2H); and ectopic glumes inside the palea/lemma (Figure 2H as compared with Figure 2C) and malformed stamens (Figure 2I) developed. In most of the florets, stamens were decreased and malformed (Figures 1D,I), miR172aOEs plants were infertile.

We further cloned AtmiR172a precursor from Arabidopsis and overexpressed it in rice. Mature AtmiR172a was the same as the OsmiR172a although their precursor differs (Supplementary Figure S2). The AtmiR172aOE lines showed elongated and distorted palea/lemma (indicated by stars) and indeterminacy of flowers, as indicated by increased number of rudimentary glumes (arrows), similar to the phenotype of OsmiR172aOEs (Supplementary Figure S3). This further illustrated the conservation of miR172.

To study the function of miR172 in detail, we further analyzed the function of miR172 targets genetically. In rice, five AP2 genes are predicted as targets of miR172⁴ (Zhu et al., 2009), and they were tentatively designated as AP2-1, AP2-2, AP2-3, AP2-4, and AP2-5, respectively, in this study. All the five AP2 genes have two AP2 DNA-binding domains in a similar mode (Supplementary Figure S4). To mimic the knock-down function of miR172, these AP2 genes were, respectively, overexpressed. 30–50 individual transgenic plants were gotten for each gene, and over 90% of them showed the respective phenotype illustrated below, which proved the successful genetic transformation.

In contrast to miR172aOEs plants, all AP2OE plants showed shortened lemma and palea, and so that reduced grain size (Figure 3A), except that AP2-5OE plants were infertile (Figures 3C,D). As a result, the 1000-grain-weight of AP2-1OE, AP2-2OE, AP2-3OE, and AP2-4OE decreased (Supplementary Figure S5A), with the correspondingly reduced starch granules (Supplementary Figures S5B–F). AP2-2OE and AP2-5OE showed the most severe abnormality in floral organs. The lemma/palea was shortened appropriately one third in AP2-2OE and more in AP2-5OE (Figures 3B,C). Furthermore, in the AP2-2OE, the lemma and palea developed inconsistently, with the palea longer than the lemma (Figure 3C), contrasting to those in the miR172OEs (Figure 1A); in the AP2-5OE plants, the lemma/palea was malformed (Figure 3D). So that miR172/AP2 regulated floral organ identity and flower determinacy, especially elongation of the lemma/palea.

Furthermore, miR172s were down-regulated in the AP2-1OE, AP2-2OE, and AP2-5OE plants (Figure 4), indicating the possibility of negative feedback regulation between these AP2s and miR172.

The OsMADS1 gene regulates floral organ identity, affecting lemma/palea development and spikelet determinacy (Jeon et al., 2000; Prasad et al., 2005; Wang K. et al., 2010). The phenotypic similarity of miR172OEs and the osmads1 mutant suggests a potential interaction between OsMADS1 and miR172.

To obviate the influence of the genetic background on the phenotype, we further constructed OsMADS1 over expression (OsMADS1OE) and RNAi transgenic plants (OsMADS1RNAi) in the ZH11 background; for each transgenic events, over 90% transformation efficiency were got. In the OsMADS1RNAi plants, the paleas and lemmas were slightly longer than those of the wild-type, and the lemma was longer than the palea (Figure 5A). Similarly, OsMADS1RNAi showed open hull and some ectopic glumes developed between whorl1 and whorl2 (Figure 5B, arrowhead). So that OsMADS1RNAi could mimic the osmads1 mutant, although to a milder degree.

All miR172s were down-regulated in the Osmads1OE plants, but up-regulated in the OsMADS1RNAi plants (Figure 5C). Expression of miR162 in these plants was basically the same (Figure 5C), indicating that the expression of miR172s was specifically modulated by OsMADS1.

MADS-box genes regulate flower organ identity by binding to the cis-regulatory elements in the target genes termed “CArG-boxes” [consensus 5′CC(A/T)₆GG3′] (Riechmann et al., 1996; Cui et al., 2010). In the 3 Kb promoter region of miR172a, miR172b, miR172c, and miR172d, there are 6, 4, 3, and 3 CArG-boxes, respectively (Supplementary Figure S6A). We carried out yeast one-hybrid and did not detect the direct binding of OsMADS1 protein to the motifs in the promoters of miR172 (Supplementary Figure S6B). However, OsMADS1 could interact with AP2-2, AP2-3, and AP2-5 in yeast two-hybrid systems (Supplementary Figure S7).

To further examine the relationship between miR172a and OsMADS1, we performed in situ hybridization of them in flower development. Before differentiation of the floral organs, both OsMADS1 and miR172a expressed highly at the floret meristem (Figures 6A,F); thereafter, miR172a gradually accumulated on all the floral organs, with higher expression in the inner stamens and carpels, and lower expression in the outer palea and lemma (Figures 6B–E). However, OsMADS1 mRNA gradually concentrated on the palea and lemma (Figures 6G–I). So that the expression region of OsMADS1 and miR172 showed complementary character, OsMADS1 might repress excessive accumulation of miR172 in the lemma and the palea.

Assessment of the expression of miR172a in the Osmads1OE and OsMADS1RNAi lines showed that during development of the empty glume and palea/lemma, miR172a was down-regulated in the OsMADS1OE lines (Figure 6J as compared with Figure 6M); during the development of inner floral organs, miR172a expressed similarly in the OsMADS1OE and OsMADS1RNAi lines (Figure 6K as compared with Figure 6N, Figure 6L as compared with Figure 6O). Therefore, during the developmental process of palea/lemma, OsMADS1 exhibited inhibition to miR172, which may further explain the longer palea/lemma of miR172OEs plants.

Since the miR172aOE, miR172cOE, and miR172dOE were totally infertile, and miR172bOE barely produced any seeds, these lines were not amenable to genetic manipulation, we used AP2 genes to perform genetic complementation.

First, we made a cross between AP2-3OE and OsMADS1RNAi plants. Among the 11 hybrids of AP2-3OE/OsMADS1RNAi in the F1 generation, six individuals showed shortened and closed hull relatively to the OsMADS1RNAi plants (Figures 7A,B), the elongated and inconsistent palea/lemma in OsMADS1RNAi plants could be rescued by AP2-3 over expression (Figures 7A,B).

Next, the AP2-2OE plasmid was genetically transformed into the OsMADS1RNAi plants. The lemma was longer than the palea in both the miR172aOE and OsMADS1RNAi plants, but shorter than the palea in the AP2-2OE plants. In the AP2-2/OsMADS1RNAi plants, the palea/lemma developed synchronously, and they were shorter than those of the OsMADS1RNAi plants but similar to those of the wild-type (Figure 7C). These results indicated the elongated and inconsistent palea/lemma in the OsMADS1RNAi plants could be rescued by AP2-2 over expression.