Lines PFG_K-03014.R of rice (Oryza sativa japonica cv. KitaaKe), which carries a T-DNA insertion in LOC_Os09g13630 (OsEMF2b), was obtained from RiceGE¹. All plants were grown in rice paddies at Wenjiang, Sichuan and Lingshui, Hainan, where represent natural long-day and short-day conditions, respectively. Genotypes were determined by PCR using a pair of gene-specific primers listed in Supplementary Table S2.

For pollen viability, pollen was stained using I₂-KI solution and observed under a Zeiss Imager A2 microscope. The panicle photograph was taken using the Canon EOS 60D digital camera and the spikelets were observed using a Zeiss Discovery V20 microscope. Anther development was classified into three stages as in Feng et al. (2000) and Zhang and Wilson (2009) and semi-thin sections were prepared as described (Qin et al., 2013). Briefly, spikelets at various developmental stages were fixed in 3% (v/v) glutaraldehyde in 0.2 M sodium phosphate buffer (pH 7.0), then dehydrated in a graded series of ethanol solutions, embedded in Technovit 7100 resin (Hereaus Kulzer, Germany), polymerized at 45°C, and sectioned at 2 μm thickness using a microtome (Leica RM2235, Germany). The sections were stained with 0.25% (w/v) toluidine blue (Sigma) and photographed using a Nikon E600 microscope.

Wild type spikelets of different developmental stages were fixed in 5% acetic acid, 50% ethanol, and 3.7% formaldehyde in DEPC water for 16 h at 4°C, then dehydrated through an ethanol series, embedded in Paraplast Plus (Sigma), and sectioned at 8 μm using a microtome (Leica RM2235, Germany). A gene-specific region of OsEMF2b was amplified for probe preparation using OsEMF2b specific primers, and cloned into a pBluescript vector (TransGen Biotech, China), then digested with EcoRI or BamHI and transcribed with T7 RNA polymerase using the DIG RNA labeling kit (Roche), for both antisense and sense probes. RNA hybridization and probe detection were performed as described (Zhu et al., 2011).

Total RNAs were extracted using TRIzol reagent (Invitrogen) from rice leaves and two stages of inflorescences, according to distance from the auricle of 0 or 1 cm, which are the standards for meiosis and early young microspore stage, respectively (Kerim et al., 2003). 1 μg total RNAs were used to synthesized first-strand cDNA using Primescript RT reagent Kit with gDNA Eraser following the manufacturer’s instructions [TaKaRa Biotechnology (Dalian), Co., Ltd, Japan]. Specific primers were designed according for individual genes (Supplementary Table S2). First-strand cDNA (2 μl) was amplified with 2 μl of specific primers in a total reaction volume of 20 μl, and each reaction was performed in triplicate using KAPA SYBR FAST qPCR Master Mix (KAPA BIOSYSTEMS) in a Thermal Cycler CFX96 (Bio-Rad, United States). The ubiquitin gene was used as an internal control for data normalization and changes in gene expression were calculated using the 2^ΔΔC_T algorithm (Liu et al., 2012). The means from three biological replicates were used for analysis.

mCherry cDNA was amplified from pBIN20-ER-rk (Nelson et al., 2007) with the primer pairs mCherry F and mCherry R, and cloned into pC2300-35S-eGFP vector (Ma et al., 2017) to generate pC2300-mCherry with KpnI and BamHI restriction enzyme sites. The coding sequence of OsEMF2b was amplified using the primer pairs OsEMF2b-mCherry F and OsEMF2b-mCherry R listed in Supplementary Table S2. The amplified fragment and pC2300-mCherry backbone were digested with BamHI and XbaI, then ligated together to create pC2300-mCherry-OsEMF2b. The pC2300-mCherry and recombinant vector pC2300-mCherry-OsEMF2b were then transformed into rice protoplasts as described (Zhang et al., 2011). mCherry fluorescence was visualized using a ZEISS LSM 710 confocal microscope.

The entire coding region of OsEMF2b was amplified with KOD DNA polymerase (TOYOBO, Japan) using the primer pairs OsEMF2b-comp F and OsEMF2b-comp R. The amplified fragment was digested with SpeI and PstI and then ligated to pCactN carrying the rice ACTIN promoter (Zhu et al., 2016) to generate pCactN-EMF2b, which was transformed into osemf2b-4 using Agrobacterium tumefaciens (Hiei et al., 1994). Positive transgenic plants were selected with G418 and identified using the primer pair for Neomycin phosphotransferase II gene (NPT II).

Statistical significance of differences in the mean values was determined using the Student’s t-test function in Microsoft Excel software. Levels of significance were represented by asterisks as follows: ^∗,^∗∗ indicates significance at P ≤ 0.05 or 0.01, respectively.

Based on RiceGE database and sequencing results, the T-DNA insertion of osemf2b-4 was inserted -200 bp upstream of the OsEMF2b start codon (Supplementary Figure S1). RT-qPCR analysis revealed that the expression of OsEMF2b was significantly decreased in leaves of osemf2b-4 (Figure 1A), indicating that it was a weaker allele than the three null mutant alleles (Kim et al., 2010). The osemf2b-4 homozygous mutant was slightly shorter than wild type (Figure 1B). The most obvious phenotypes of osemf2b-4 were reduced grain number and seed set (Figure 1C). The grain number per panicle of wild type was 59 ± 3.6, whereas osemf2b-4 was 22.3 ± 2.1. Additionally, a variety of abnormal seeds were observed, for example, some seeds protruded from the hull, and some had a long sterile lemma (Figure 1D).

To test whether the pleiotropic phenotype of osemf2b-4 was caused by the T-DNA insertion in its promoter region, we firstly carried out co-segregation analysis using T₁ and T₂ segregation populations (Supplementary Table S1). The coding sequence of OsEMF2b driven by rice ACTIN promoter was introduced into homogenous osemf2b-4, and 4 T₀ transgenic plants were obtained (Supplementary Figure S2), the 3 with highly enhanced expression of OsEMF2b exhibited normal florets, anther and pollen, whereas the one with relatively low expression still exhibited defective floret, anther and pollen development (Supplementary Figures S2A–C, S3A–J). Together, these results suggested that the T-DNA insertion in OsEFM2b promoter region was responsible for the osemf2b-4 phenotypes.

Previous studies on biological and molecular function of OsEMF2b were focused on flowering time, inflorescence and floret development using its null mutant (Luo et al., 2009; Conrad et al., 2014; Xie et al., 2014). Here, we used the weak allele osemf2b-4 to study its function during post-meiotic anther and pollen development. The spikelets of osemf2b-4 exhibited various morphologies of palea and/or lemma (Figure 2A). We divided these florets with stamen into two different types, slight defects (Figures 2B–E) and severe defects (Figures 2a–d). In osemf2b-4, there was a high percentage of slightly defective types including small husk, beak-like, open husk and long sterile lemma florets types, 34.5, 34.5, 13.8, and 6.9% (n = 142), respectively. The most severe cases, such as double lemma (3.4%, n = 142), coiled lemma and palea (3.4%, n = 142), missing lemma or palea (2%, n = 142), and leaf-like floret (1.4%, n = 142).

The palea and lemma were removed for clearly visualizing anthers (Figures 2F–J,e–h). In osemf2b-4, normal anthers were observed in most florets, but the number of anthers in each floret were variable. In the three null mutants anthers were rarely produced but did not contain any pollen grains (Luo et al., 2009; Kim et al., 2010). I₂-KI staining showed that the pollen fertility of osemf2b-4 anthers was variable, from semi-sterility to complete sterility (Figures 2L–O,i–l), when compared with the fertility of WT anthers (Figure 2K). Abortive pollen grains were observed in the transparent and white stamens (Figures 2P–R). These results suggested that the partial spikelet sterility of the osemf2b-4 mutant was likely caused by reduced pollen viability, and that osemf2b-4 was an excellent allele for studying the function of OsEMF2b during post-meiotic anther and pollen development.

To more precisely determine the cytological reason for the post-meiosis anther and pollen defects in osemf2b-4, we performed semi-thin sections for at three stages of anther development (Feng et al., 2000; Zhang and Wilson, 2009). According to the anther defects in different locules, osemf2b-4 anthers were grouped into three different categories from young microspore stage to mature pollen stage, respectively (Figures 3A–L). At the young microspore stage, the tapetal cells was deeply stained, and exine was deposited on the round microspores in wild type anthers (Figure 3A). However, three different phenotypes were observed in osemf2b-4 anthers: type I (74%, n = 45) with likely normal tapetum and irregular shape of microspores (Figure 3B); type II (20%, n = 45) with lightly stained and weakly expanded tapetum and absent middle layer (Figure 3C); type III (6%, n = 45) with extremely enlarged and disordered tapetum and no exine on microspores (Figure 3D). At the vacuolated stage in wild type anthers, the tapetal cells had become more degenerated and the microspores appeared vacuolated (Figure 3E). The mutant anthers exhibited three different types, including type I (16%, n = 146) with likely normal locules (Figure 3F), type II (17%, n = 146) with shrunken microspores (Figure 3G), and type III (67%, n = 146) with delayed degradation of tapetum and folded microspores (Figure 3H).

By the mature pollen stage, the tapetum and middle layer had completely degenerated and the endothecial cell layer eventually was disrupted to release pollen grains in wild type (Figure 3I). In osemf2b-4, 32% of anther locules (n = 38) displayed linear shaped surface, similar to wild type anthers. Some pollen grains were stained weakly with toluidine blue, while some grains were irregularly shaped and collapsed in type I (Figure 3J). Another two types of anther defects, type II (45%, n = 38) and type III (23%, n = 38) exhibited glossy and smooth anther surface and shrunken pollen, as well as delayed degradation of the endothecial cell layer (Figures 3K,L). Together, these observations suggested that osemf2b-4 had developmental defects in post-meiotic anther wall layers and microspores, and that OsEMF2b plays an essential role in post-meiotic anther and pollen development.

Given the anther defects observed in osemf2b-4, we examined the expression pattern of OsEMF2b at two developmental stages using in situ hybridization (Figures 4A–C). OsEMF2b transcripts were detected in microsporocyte, tapetum and vascular bundle of anther connective at the meiosis stage (Figure 4A). OsEMF2b had high expression in tapetum and weak expression in microspores and vascular bundle at the young microspore stage (Figure 4B). Only background level of signal was detected with the sense probe (Figure 4C).

Even though OsEMF2b was predicted to be localized in the nucleus (Yoshida et al., 2001; Luo et al., 2009), due to two putative nuclear localization signals, a single C₂H₂-type zinc finger motif and a VRN2-EMF2-FIS2-SUZ12 (VEFS) domain (Figure 5A), no experimental data showed its protein localization. In order to test the subcellular localization of OsEMF2b, a 35S::mCherry-OsEMF2b fusion construct was introduced into rice leaf protoplasts. As expected, the mCherry-OsEMF2b fusion protein was observed in the nucleus (Figures 5Bd–f), whereas the empty vector with 35S::mCherry was observed in cytoplasm (Figures 5Ba–c).

Defective anther development was observed in osemf2b-4, probably by affecting the gene expressions involved in post-meiotic anther and pollen development. We used the auricle distance between the penultimate leaf and flag leaf for assessing anther development stage; 0 and 1 cm respectively represented meiosis and young microspore stage (Kerim et al., 2003). Then we analyzed the expression of several nuclear genes involved in anther and pollen development at these two stages. Most anther and pollen-related genes had tissue-specific expression patterns. TAPETUM DEGENERATION RETARDATION (TDR) (Li N. et al., 2006) and ETERNAL TAPETUM1 (EAT1) (Niu et al., 2013) were preferentially expressed in the tapetum, as HISTONE MONOUBIQUITINATION1/2 (OsHUB1/2) (Cao et al., 2015), GAMYB (Aya et al., 2009), APPOTOSIS INHIBITOR5 (OsAPI5), OsAIP1/2 (Li X. et al., 2011), CARBON STARVED ANTHER (CSA) (Zhu et al., 2015), MYB IMPORTANT FOR DROUGHT RESPONSE1 (MID1) (Guo et al., 2016) and PERSISTENT TAPETAL CELL1 (PTC1) (Li H. et al., 2011) were expressed in the tapetum and microspores. OsRH2/34 (Huang et al., 2016), OsMYB106 and OsSPX1 (Zhang et al., 2016) were also essential for anther and/or pollen development. PTC1 expression was significantly decreased in osemf2b-4, and the other 14 genes were significantly up-regulated at stage 2 (Figure 6), suggesting that OsEMF2b was required for fine-tuning the expression of these genes involved in the anther and pollen development.