The bHLH142 Transcription Factor Coordinates with TDR1 to Modulate the Expression of EAT1 and Regulate Pollen Development in Rice.

This study shows that rice bHLH142 functions in tapetal programmed cell death and pollen development, finding that bHLH142 and TDR1 form a heterodimer that binds to the EAT1 promoter to activate its expression. Protein modeling also suggests that EAT1 can form a heterodimer with TDR1, presumably serving as a feedback regulator. Male sterility plays an important role in F1 hybrid seed production. We identified a male-sterile rice (Oryza sativa) mutant with impaired pollen development and a single T-DNA insertion in the transcription factor gene bHLH142. Knockout mutants of bHLH142 exhibited retarded meiosis and defects in tapetal programmed cell death. RT-PCR and in situ hybridization analyses showed that bHLH142 is specifically expressed in the anther, in the tapetum, and in meiocytes during early meiosis. Three basic helix-loop-helix transcription factors, UDT1 (bHLH164), TDR1 (bHLH5), and EAT1/DTD1 (bHLH141) are known to function in rice pollen development. bHLH142 acts downstream of UDT1 and GAMYB but upstream of TDR1 and EAT1 in pollen development. In vivo and in vitro assays demonstrated that bHLH142 and TDR1 proteins interact. Transient promoter assays demonstrated that regulation of the EAT1 promoter requires bHLH142 and TDR1. Consistent with these results, 3D protein structure modeling predicted that bHLH142 and TDR1 form a heterodimer to bind to the EAT1 promoter. EAT1 positively regulates the expression of AP37 and AP25, which induce tapetal programmed cell death. Thus, in this study, we identified bHLH142 as having a pivotal role in tapetal programmed cell death and pollen development.


INTRODUCTION
Rice (Oryza sativa) is one of the most important staple crops in the world, feeding almost half of the world's population, and it serves as a model for monocots, which include many important agronomic crops (e.g., wheat [Triticum aestivum], maize [Zea mays], sorghum [Sorghum bicolor], and millet [Setaria italica]). The Food and Agriculture Organization of the United Nations predicts that rice yield will have to be increased 50 to 70% by 2050 to meet demands. Several approaches are currently adopted to increase rice yields, such as heterosis breeding, population improvement, wide hybridization, genetic engineering, and molecular breeding (Khush, 2000). Among these, hybrid rice is considered the most promising strategy, producing 15 to 20% increases in yield (Gao et al., 2013). Male sterility plays an important role in the development of hybrid crops, especially self-pollinated crop species. Male-sterile traits can be divided into cytoplasmic male sterility (CMS), which is determined by cytoplasmic factors such as mitochondria, and genetic male sterility (GMS), which is determined by nuclear genes. CMS has long been used in hybrid maize production, and both CMS and GMS are currently used for hybrid rice production (Luo et al., 2013). In the case of CMS-based hybrid crop production, a maintainer line is needed for the maintenance of the male sterile line.
Rice anthers are composed of four lobes attached to a central core by connective and vascular tissue. When anther morphogenesis is completed, microsporocytes form in the middle, surrounded by four anther wall layers: an epidermal outer layer, endothecium, middle layer, and tapetum (Goldberg et al., 1993). The tapetum is located in the innermost cell layer of the anther walls and plays an important role in supplying nutrients such as lipids, polysaccharides, proteins, and other nutrients for pollen development (Zhu et al., 2008). The tapetum undergoes programmed cell death (PCD) during the late stage of pollen development (Wu and Cheun, 2000); this PCD causes tapetal degeneration and is characterized by cellular condensation, mitochondria and cytoskeleton degeneration, nuclear condensation, and internucleosomal cleavage of chromosomal DNA. Tapetal PCD must occur at a specific stage of anther development for normal tapetum function and pollen development, and premature or delayed tapetal PCD and cellular degeneration can cause male sterility (Papini et al., 1999;Kawanabe et al., 2006;Li et al., 2006a;Luo et al., 2013).
Genetic and functional genomic studies of male sterility in Arabidopsis thaliana have shown that many transcription factors (TFs) play essential roles in pollen development and the regulation of tapetal PCD, such as mutations in DYSFUNCTIONAL TAPETUM1 (DYT1), DEFECTIVE IN TAPETAL DEVELOPMENT AND FUNCTION1 (TDF1, MYB35), ABORTED MICROSPORES (AMS), and MALE STERILITY1 (MS1); mutations in these factors all result in a male-sterile phenotype. The genetic regulatory pathway of pollen development suggests that DYT1 (Zhang et al., 2006), TDF1 (Ito and Shinozaki, 2002), and AMS (Sorensen et al., 2003) function at the early stage of tapetum development, while AtMYB103/80 (Higginson et al., 2003;Phan et al., 2011) and MS1 (Wilson et al., 2001;Ito and Shinozaki, 2002;Ito et al., 2007) play important roles in late tapetum development and pollen wall formation. In rice, several TFs, such as UNDEVELOPED TAPETUM1 (UDT1; homolog of DYT1), are known to be key regulators of early tapetum development (Jung et al., 2005). In addition, mutations in TAPETUM DEGERATION RETARDATION (TDR1) (Li et al., 2006a), GAMYB (Aya et al., 2009;Liu et al., 2010), ETERNAL TAPETUM1 (EAT1) (Niu et al., 2013), and DE-LAYED TAPETUM DEGENERATION (DTD) (Ji et al., 2013) all cause male sterility associated with tapetal PCD. TDR1, an ortholog of Arabidopsis AMS, plays an essential role in tapetal PCD in rice, and tdr1 shows delayed tapetal degeneration and nuclear DNA fragmentation as well as abortion of microspores after release from the tetrad. Molecular evidence indicates that TDR1 directly binds the promoters of CP1 and C6 to regulate their transcription (Li et al., 2006a). CP1 is involved in intercellular protein degradation and its mutant shows defects in pollen development (Lee et al., 2004), and C6 encodes a lipid transfer protein that plays a crucial role in the development of lipidic orbicules and pollen exine during anther development (Zhang et al., 2010a). EAT1 acts downstream of TDR1 and directly regulates the expression of AP25 and AP37, which encode aspartic proteases involved in tapetal PCD (Niu et al., 2013). Microarray analysis of the coexpression gene networks during rice pollen development suggest that 108 genes may be involved in pollen wall synthesis and predicted that Os01g0293100 (bHLH142 in this study) is directly connected to the expression of CYP703A3, CYP704B2, MS2, and C6, which may function in sporopollenin biosynthesis (Aya et al., 2011). Consistent with this, cyp704B2 mutants showed swollen sporophytic tapetal layers, aborted pollen grains without exine, and undeveloped anther cuticles (Li et al., 2010).
The basic helix-loop-helix (bHLH) proteins are a superfamily of TFs and one of the largest TF families in plants. There are at least 177 bHLH genes in the rice genome (Li et al., 2006b;Carretero-Paulet et al., 2010) and more than 167 bHLH genes in the Arabidopsis genome (Bailey et al., 2003;Toledo-Ortiz et al., 2003). Generally, eukaryotic TFs consist of at least two discrete domains, a DNA binding domain and an activation or repression domain that operate together to modulate the rate of transcriptional initiation from the promoter of target genes (Ptashne, 1988). The bHLH TFs play many different roles in plant cell and tissue development as well as in plant metabolism. The HLH domain promotes protein-protein interaction, allowing the formation of homodimeric or heterodimeric complexes (Massari and Murre, 2000). They bind as dimers to specific DNA target sites and are important regulatory components in diverse biological processes . So far, three bHLH TFs have been shown to be involved in rice pollen development: UDT1 (bHLH164) (Jung et al., 2005), TDR1 (bHLH5) (Li et al., 2006a), and EAT1/DTD1 (bHLH141) (Ji et al., 2013;Niu et al., 2013).
By screening a T-DNA-tagged rice mutant pool of TNG67 (Hsing et al., 2007), we isolated a male sterility-related nuclear gene encoding a member of the bHLH TFs (bHLH142). In this study, we elucidate the molecular mechanisms underlying male sterility in this mutant. Our molecular and protein modeling results suggest that bHLH142 is specifically expressed in the anther and bHLH142 coordinates with TDR1 in regulating EAT1 promoter activity in transcription of protease genes required for PCD during pollen development. Therefore, we assert that bHLH142 plays an essential role in rice pollen development by regulating tapetal PCD.

Identification of a Male-Sterile Rice Mutant
From the T2 population of the Taiwan Rice Insertional Mutants (TRIM) lines, we identified a T-DNA-tagged rice mutant (denoted ms142) with a completely male-sterile phenotype. In the field, this mutant produced no viable seeds but maintained normal vegetative growth ( Figure 1A) with panicles and spikelets developing similarly to those of the wild type ( Figures 1B to 1D). The ms142 mutant exhibits normal opening of spikelets and elongation of anther filaments ( Figures 1D). However, the anthers of ms142 were significantly smaller in size and appeared yellowish white ( Figure 1D). The anthers of ms142 could not be stained by Sudan black due to lack of lipid accumulation ( Figure  1E) and showed no pollen grain development ( Figure 1G).

Sequence Analysis of the T-DNA-Tagged Gene in the ms142 Mutant
To determine T-DNA insertion copy number, DNA gel blot analysis of T2 mutant lines was conducted using hptII as a probe. Only a single band was detected in the mutant lines (Supplemental Figure 1A); thus, the mutation in ms142 is due to a single T-DNA insertion. Analysis of the flanking sequence tag in the TRIM database suggests that ms142 is a putative mutant with T-DNA inserted 1257 bp from the ATG start codon in the 3rd intron of bHLH142 (RAP locus Os01g0293100, MSU locus Os01g18870). The protein encoded by this gene is annotated as a basic helixloop-helix dimerization region bHLH domain-containing protein (RiceXPro, version 3.0). The bHLH142 gene has four exons and three introns. Furthermore, genotyping by PCR using specific primers crossing the T-DNA insertion site verified its flanking sequence tag (Supplemental Figure 1).

Agronomic Traits of the ms142 Mutant and Genetic Study
The agronomic traits of the mutant were examined in the selfed progenies of heterozygous mutants grown in the outdoor screen house. Heterozygous plants behaved similarly to the wild type in terms of vegetative and reproductive growth and produced fertile seeds. However, homozygous ms142 mutant plants exhibited similar plant height, panicle number, and panicle length to the wild type but produced no viable seeds (Supplemental Table 1).
To understand whether the sterility in ms142 is due to male or female sterility, homozygous mutants were backcrossed with wild-type pollen. All F1 plants displayed wild-type-like growth and fertility. These results imply that the female organs of ms142 develop normally. When the ms142 BCF1 was selfed, the BCF2 progeny segregated into fertile and sterile plants in a ratio of 3:1 (Supplemental Table 2), suggesting that the male-sterile trait is controlled by a recessive gene. Consistent with the mutant phenotype, backcross segregants showed male sterility only in the homozygous plants, indicating that the male sterile phenotype cosegregated with the genotype. Moreover, when the selfed seeds derived from heterozygous plants of T2, T3, T4, and BCF2 generations were planted in different years and different cropping seasons, the scoring of the phenotype indicated that male sterility in ms142 is stable and not affected by cropping season or year. Again, the fertile and sterile plants segregated in an ;3:1 ratio, as supported by x 2 analysis (data for T4 and BCF2 shown in Supplemental Table 2). Taken together, these genetic analyses confirmed that the male sterility in ms142 is controlled by a single recessive locus.

Defects in Anther Wall and Pollen Development in the ms142 Mutant
To investigate the defects in the anthers of ms142, we examined the anatomy of anthers in the wild type and homozygous mutant. At the microspore mother cell (MMC) stage, the wild-type anther walls contained an epidermal cell layer, endothecial cell layer, middle layer, and tapetal cell layer ( Figure 2A). During the early meiosis stage, the MMCs underwent meiosis to form tetrads of haploid microspores, the tapetal cells differentiated to form a large vacuole, and the middle layer cells began to degenerate ( Figure 2B). At the tetrad stage, the meiocytes formed tetrads ( Figure 2C). During the young microspore stage, free microspores were released into the anther locule, and the microspores developed and exine was deposited on the pollen grain wall. The middle layers shrank and the tapetal cell layers became very dense ( Figure 2D). At the mature pollen stage, the uninucleate pollen developed to trinucleate pollen with starch, protein, lipid, and other nutrients enriched in the pollen cytoplasm. At maturity, the tapetal cells were completely degenerated and the endothecial cell layers were thickening, ready for anther dehiscence ( Figure 2E).
At the MMC stage, there were no visible differences in the anthers between the wild type and ms142. The ms142 anther consisted of normal epidermis, endothecium, middle layer, and tapetum ( Figure 2F). During the early meiosis stage, however, ms142 MMCs did not enter meiosis and formed abnormal organelles ( Figure 2G, indicated by arrow). Abnormal endoplasmic reticulum structure and apoptosis was also observed by transmission electron microscopy ( Figure 2N, indicated by arrow). The ms142 tapetal cells continuously became vacuolated and elongated, with some cells divided into two tapetal layers ( Figure  2G, indicated by arrowheads). Transmission electron microscopy clearly showed two tapetal cells developed in the mutant ( Figure 2L, indicated by arrowheads). The mid-layers of the mutant tapetum maintained their initial shapes, but meiocytes failed to divide into four cells at the tetrad stage ( Figure 2H). The ms142 mutant microspores finally degenerated during the vacuolated pollen stage ( Figure 2I, indicated by asterisk). The tapetal and middle layer cells contained a large vacuole, and the middle layer cells did not degenerate ( Figure 2I). Consequently, there were no mature pollen grains formed in the locules at the mature stage. The mutant anther wall still retained four to five layers of cells, i.e., epidermis, endothecium, middle layer, and one or two layers of tapetum cells. By contrast, the endothecial cell layer did not become thickened in the mutant even at the latter stage of anther development ( Figure 2J).

Meiotic Activity in the ms142 Mutant
Since early meiosis was arrested in ms142 (Figure 2), we further performed meiotic analysis using acetocarmine to stain MMC chromosomes before meiosis (Chang et al., 2009b). The mutant exhibited no dyad development and meiocytes degenerated during meiosis; consequently, no microspore development was found in the mutant (Supplemental Figure 2A). By contrast, the MMCs of the wild type underwent normal meiosis and formed tetrad with four haploid cells and further developed into normal microspores.  To find out what caused retardation of meiocyte development in ms142, we compared the transcript levels of some meiosis regulatory genes by RT-PCR. The results indicated that Mel1, which functions in premeiotic germ cell division (Nonomura et al., 2007), was not significantly altered in ms142, while PAIR1 (Nonomura et al., 2004), PAIR3 (Yuan et al., 2009), and SDS (Chang et al., 2009a), important for chromosome pairing or synapsis during meiosis I, did not show any significant difference in expression between wild type and ms142 (Supplemental Figure 2B). Therefore, the male-sterile phenotype in the mutant is not related to meiosis.

Mutated bHLH142 Causes Defects in Tapetal PCD
Histological analysis indicated that ms142 has abnormal anther morphology and aborted development of tapetal cells (Figure 2), which is not due to alteration in expression of meiosis-related genes (Supplemental Figure 2B). Thus, we suspected that mutation of bHLH142 might cause altered tapetal PCD and thus affect tapetal degeneration (Papini et al., 1999). Tapetal PCD is characterized by cellular condensation, mitochondria and cytoskeleton degeneration, nuclear condensation, and internucleosomal cleavage of chromosomal DNA (Phan et al., 2011). Therefore, we performed a terminal deoxynucleotidyl transferase-mediated dUTP nick-end labeling (TUNEL) assay to detect DNA fragmentation in the anthers of the wild type and ms142. A TUNEL-positive signal began to appear in the tapetal cells of the wild type during meiosis and a strong TUNEL signal was detected during the young microspore stages ( Figure 3). In contrast, no DNA fragmentation was observed in the tapetal layer in ms142 throughout anther development ( Figure 3). Differential interference contrast images of anther cross sections corresponding to TUNEL sections are shown in Supplemental Figure 3.
bHLH142 Is a Nuclear Protein The gene structure of bHLH142, shown in Supplemental Figure  4A, indicates that the bHLH domain contains a bipartite nuclear localization signal (NLS), and the gene is predicted to encode a protein of 379 amino acids with a theoretical molecular mass of 40.7 kD and a pI of 6.2 (http://web.expasy.org/ compute_pi). The predicted 3D protein structure of bHLH142 clearly shows the bHLH domain comprises two a-helices (Supplemental Figure 4C). Since the bHLH proteins are characterized as TFs, we assumed that bHLH142 is localized in the nucleus. To verify the subcellular localization, we constructed a fusion gene of the green fluorescent protein gene (GFP) and bHLH142 under the control of the 35S promoter and the nos terminator for transient expression in rice leaf mesophyll protoplasts (Supplemental Figure 4D). As a positive control, the NLS sequence was also fused to the red fluorescent protein (RFP) gene using the same regulatory elements. These constructs were introduced into rice protoplasts, and as expected, with the GFP construct alone, free GFP was found in the nucleoplasm as well as in the cytoplasm. However, the bHLH142: GFP fusion protein and the positive control of NLS:RFP were exclusively located in the nucleus (Supplemental Figure 4E). These results confirmed that, as a TF, bHLH142 protein localizes in the nucleus. DNA fragmentation signals (yellow fluorescence) started at the tetrad stage and exhibited obvious positive signals at the young microspore stage in TNG67. No DNA fragmentation signal was observed in the tapetum of ms142. The red signal is propidium iodide staining, and the yellow fluorescence is the merged signal from TUNEL (green) and propidium iodide (red). Bars = 50 mm.

Phylogenetic Analysis
To understand the evolution of bHLH142 among various organisms, we used full-length bHLH142 protein sequence to search the National Center for Biotechnology Information BLAST database and retrieved 21 homologs containing the bHLH domain from 10 diverse terrestrial plants. The phylogenetic tree shows that UDT1 (bHLH164) and TDR1 (bHLH5) are in the same cluster, while rice bHLH142 and EAT1 (bHLH141) evolved and diversified into two separate clades. Phylogenetic analysis also suggests that bHLH142 is descended from a common monocot ancestor. Rice bHLH142 shares high sequence similarity with related proteins from Brachypodium distachyon, millet, Triticum urartu, maize, sorghum, and Aegilops tauschii (Supplemental Figure 5). The conserved homologs of bHLH142 from major cereal crops, such as maize, millet, sorghum, and wheat, share 84.1, 79.2, 72.2, and 78% amino acid sequence similarity to their rice counterpart (Supplemental Table 4). The maize homolog GRMZM2G021276 (ZmLOC100283549) is highly expressed in immature tassel, meiotic tassel, and anther (http://bar.utoronto. ca/efp_maize/cgi-bin/efpWeb.cgi). In agreement, our RT-PCR data also verified that the maize homolog is tissue-specifically expressed in meiotic anther (data not shown). This result implies that the maize bHLH142 homolog may also play a similar role in anther and pollen development.

Expression Pattern of bHLH142
RT-PCR and quantitative RT-PCR (qRT-PCR) analyses of the wild type showed that bHLH142 mRNA accumulates in young rice panicle and anther only, and not in other tissues (e.g., root, shoot, leaf, lemma, palea, ovary, and seed). In particular, high levels of transcripts were found in developing panicles (Figures 4A and 4B). Specifically, bHLH142 transcripts were highly expressed in MMCs and extremely highly expressed in the anther at the meiosis stage, but this expression gradually decreased after the microspore stage ( Figure 4C). Also, in situ hybridization (ISH) clearly demonstrated the specific expression of bHLH142 in the anther but not in lemma and palea of wild-type spikelet at the early meiosis stage ( Figure 4D). ISH with the cross sections of wild-type anther at various developmental stages showed positive signals in the tapetal layers at the early meiosis stage and in the tapetal layers and meiocytes during meiosis stage, with decreasing signals at the young microspore stage and negligible signals after the vacuolated pollen stage ( Figure 5). Interestingly, the ISH signal was also detected in the vascular bundle ( Figure 5), suggesting that the target genes of bHLH142 might also be associated with nutrient acquisition in the anther. Conversely, there was no ISH signal detected in the anthers of the null mutant of ms142 ( Figure 5).
In addition, the expression patterns of various known pollen regulatory genes in the anther of the wild type versus ms142, as examined by qRT-PCR, confirmed the knockout of the bHLH142 transcript in the ms142 null mutant ( Figure 6A). Also, expression of TDR1, EAT1 (bHLH141), AP37, AP25, CP1, CYP703A3, CYP704B2, MS2, and C6 was significantly downregulated in the ms142 anther, relative to the wild-type anthers ( Figures 6E to  6M). However, MSP1 and UDT1 transcripts were upregulated in the mutant at the MMC and meiosis stages ( Figures 6B and 6C). There was no significant change in the GAMYB transcripts in ms142 ( Figure 6D). Interestingly, the extent of the suppression of TDR1 expression ( Figure 6E) in ms142 was less than those of the other downstream genes ( Figures 6F to 6M). To verify the gene hierarchy of bHLH142, we performed genetic study of udt1 (TRIM), eat1 (Tos17 mutant 'HA0530,' NIAS), and ms142. Our qRT-PCR results indicated that bHLH142 is downregulated in the anther of eat1 and ms142. Interestingly, knockout UDT1 exhibited high accumulation of bHLH142 in the early stage of MMCs. Taken together, these data suggest that bHLH142 is located downstream of UDT1 but upstream of EAT1 (Supplemental Figure 6). In the eat1 mutant, the expression of CP1 was largely downregulated, while the expression of C6 was not altered, suggesting that EAT1 may positively regulate the expression of CP1 in addition to AP37 and AP25 (Supplemental Figure 6D).

bHLH142 and TDR1 Coordinately Regulate EAT1 Promoter Activity
Based on the alterations in expression of known pollen regulatory genes in various mutants ( Figure 6; Supplemental Figure 6), we assumed that bHLH142 might regulate EAT1 promoter activity. Therefore, we performed transient promoter assays with the EAT1 pro -Luc construct. Our results demonstrated that bHLH142 or TDR1 protein alone cannot independently drive the expression of EAT1 pro -Luc. However, when combined, these two TF proteins together significantly increased Luc expression from the EAT1 promoter by up to 30-fold. However, additional expression of EAT1 in the same cells reduced Luc expression from a 30-fold down to an 18-fold increase, presumably due to the competition between EAT1 and bHLH142 in binding to TDR1 (Figure 7). These results suggest that TDR1 and bHLH142 coregulate the activity of the EAT1 promoter.
Protein Interactions between bHLH142, TDR1 (bHLH5), and EAT1 (bHLH141): Molecular Analysis Next, we performed yeast two-hybrid (Y2H) analysis to determine whether bHLH142, as bait, interacts with the prey, TDR1 or EAT1. It was previously reported that full-length EAT1 and TDR1 proteins possess self-activation activity (Ji et al., 2013;Niu et al., 2013); our Y2H study also confirmed this phenomenon (Supplemental Figure 7A). Therefore, we also constructed a truncated EAT1 Δaa(1-254) (truncated EAT1 at amino acids 1 to 254) and TDR Δaa(1-344) (truncated TDR amino acids at 1 to 344) to eliminate self-activation (Supplemental Figure 7A). Our results showed that bHLH142 is not self-activated (Supplemental Figure 7A); only the yeast strains coexpressing both bHLH142 and TDR Δaa(1-344) grew normally on stringent selection media ( Figure 8A), and there was no direct interaction between bHLH142 and EAT1 Δaa(1-254) . Thus, bHLH142 did not directly interact with EAT1 ( Figure 8A), and the retention of the C-terminal sequences of TDR1 is sufficient to confer the interaction of the two proteins. Clearly, the amino acid sequences in TDR Δaa(1-344) and EAT1 Δaa(1-254) contain the interacting sites, consistent with the result of Ji et al. (2013). These results are further supported by the results of EAT1 promoter assays in that both bHLH142 bHLH142 Regulates Pollen Development in Rice and TDR1 are required for the transcription of EAT1 (Figure 7). Moreover, bimolecular fluorescence complementation (BiFC) assays showed that yellow fluorescent protein (YFP) signals are detected only in the nucleus of the rice cells coexpressing both NYN1-bHLH142 and CYN1-TDR1 ( Figure 8B) and in the cells coexpressing both NYN1-TDR1 and CYN1-EAT1 ( Figure 8B), but not in the cells coexpressing both NYN1-bHLH142 and CYN1-EAT1 (Supplemental Figure 7B). In vitro interaction of bHLH142 and TDR1 proteins was further validated by coimmunoprecipitation (co-IP), where interaction between hemagglutinin (HA) fused TDR1 and bHLH142 was confirmed ( Figures 8C and 8D). Taken together, these molecular data provide solid evidence of the physical interaction between TDR1 and bHLH142.
Protein Interactions between bHLH142, TDR1 (bHLH5), and EAT1 (bHLH141): Protein Modeling Molecular dynamic (MD) simulation was conducted to determine the protein-protein interaction between bHLH142 and TDR1 and to show how the heterodimer triggers the transcription of EAT1 by binding to its promoter. The 3D structures of bHLH142, TDR1, and EAT1 were first obtained by ab initio protein modeling methods through MD simulation in an aqueous environment (Nelson et al., 1996). These proteins possess electrical and hydrophobic interactive surface patches for protein-protein interaction ( Figures 9A and 9B). As shown in Figure 9C, bHLH142 interacts with TDR1 via these patches in the C-terminals to form an bHLH142/TDR1 heterodimer. The amino acid patches of 39 Asp-47 Trp, 236 Val-245 His, and 414 Met-462 Thr on TDR1 were predicted to bind to the corresponding patches of 19 Val-23 His, 38 Phe-66 Tyr, and 85 Gln-92 His on bHLH142 ( Figure 9A) with a binding free energy of 23749 6 30 kcal/mol. Furthermore, our modeling also demonstrated that EAT1, like bHLH142, binds to TDR1 at similar sites in the C-terminal regions ( Figure 9C), with a binding free energy of 23325 6 231 kcal/mol. Specifically, the amino acid patches of 41 Trp-47 Trp, 232 Ser-246 Ala, and 414 Met-460 Val on TDR1 were predicted to bind to the corresponding patches of 168 Leu-176 Ala, 198 His-220 Pro, 274 Arg-282 Arg, and 309 Glu-319 Ile on EAT1 ( Figure 9B). Again, EAT1 interacts with TDR1 to form an EAT/TDR1 heterodimer as bHLH142 does with TDR1, with a similar interface.
The bHLH protein surface grooves of bHLH142, TDR1, and EAT1 are rich in positively charged amino acids, and the grooves on the protein surface are possible regions for binding to the promoter of target genes. As both bHLHL142 and EAT1 are required to trigger the expression of EAT1 (Figure 7), we analyzed the EAT1 promoter sequence to identify the possible DNA binding sites. A total of 14 potential E-box binding sites (CANNTG) occur in the 2-kb promoter and 59 untranslated regions (UTRs) of EAT1. One of these E-box cis-acting elements located on the promoter has a palindromic sequence (CAATTG) and is predicted to be the most favorable for binding to the dimer as it has a low free energy (Stormo and Fields, 1998). Our modeling predicted that the bHLH142/TDR1 heterodimer forms a DNA binding pocket along

RNA Interference Transgenic Rice Lines Confirmed the Role of bHLH142 in Pollen Development
To further validate the biological function of bHLH142, we generated an RNA interference (RNAi) construct to suppress the expression of bHLH142 in rice. The gene-specific region from the 59 UTR of bHLH142 was amplified, fused with b-glucuronidase (GUS) intron, and introduced into wild-type calli via Agrobacterium tumefaciens. All 16 T0 RNAi transgenic lines obtained had a male sterile phenotype similar to the T-DNA mutant ms142. These RNAi lines showed reduced expression of bHLH142, as examined by RT-PCR, and produced poorly developed anthers Error bars indicate the SD of three biological replicates. Mei, meiosis; VP, vacuolated pollen. without pollen grains (Supplemental Figure 8). This result further supports the notion that bHLH142 plays a key role in rice anther and pollen development.

DISCUSSION bHLH142 Is a Major Regulator of Rice Anther Development
Previously, three bHLH TFs were shown to be involved in pollen development in rice, and mutations of these TF genes all lead to complete male sterility. These genes are UDT1 (bHLH164) (Jung et al., 2005), TDR1 (bHLH5 (Li et al., 2006a), and EAT1/DTD1 (bHLH141) (Ji et al., 2013;Niu et al., 2013). They all play important roles in pollen development by regulating tapetal PCD. In this work, we identified a rice male-sterile mutant, ms142, with T-DNA inserted in the intron of bHLH142, which encodes another bHLH domain-containing TF protein. The bHLH142 mutant phenotype is characterized by small anthers without pollen grain development (Figure 1). Genetic analyses suggested that the mutation is due to a single T-DNA insertion event. We further showed that this TF is located in the nucleus (Supplemental Figure 4) and plays an essential role in regulating rice pollen development. Close anatomical examination of anther development (Figure 2), TUNEL assay of DNA fragmentation (Figure 3), ISH gene localization (Figure 5), and expression of key gene transcripts ( Figure 6) in the null mutant demonstrated that defects in microspore development are associated with defected tapetal PCD. Timely degradation of tapetum cells is essential for viable pollen development. Furthermore, suppressed expression of bHLH142 in wild-type rice by RNAi confers the male-sterile phenotype (Supplemental Figure 8). Thus, this study shows the involvement of the bHLH TF bHLH142 in the dynamic regulation of pollen development in rice and likely in other plants as well.
Our analysis of expression profiles of known regulatory genes involved in pollen development demonstrated the downregulation of several genes, such as TDR1, EAT1, AP37, AP25, CP1, CYP703A3, CYP704B2, MS2, and C6 etc., in ms142 during pollen development ( Figure 6). Therefore, we suggest that bHLH142 participates upstream in the same regulatory pathway of anther development. A previous study reported that TDR1 positively (A) Schematic diagrams of the reporter, effector, and internal control plasmids used in the transient transactivation assay in rice leaf protoplasts. The reporter plasmid contains the CaMV35S minimal promoter and the EAT1 promoter sequence (2 kb) fused to the firefly luciferase gene (Luc). In the effector plasmids, bHLH142, TDR1, and EAT1 genes are under the control of the CaMV35S promoter. Nos and t35s denote the terminators of nopaline synthase and CaMV35S, respectively. The pBI221 vector contains a CaMV35S promoter driving the expression of GUS as the internal control. (B) Transactivation of the Luc reporter gene by bHLH142 and TDR in rice protoplasts. Different effectors were cotransfected with the reporter and internal control plasmid (pBI221). The data represent means of three independent transient transformations. Error bars indicate SD. Transient transformation without the effector plasmid (mini35p) was used as a negative control.
bHLH142 Regulates Pollen Development in Rice regulates CP1 and C6 (Li et al., 2006a), and a more recent study suggested that EAT1 regulates the expression of two aspartic protease genes, AP25 and AP37 (Niu et al., 2013). However, our data suggest that EAT1 may also directly regulate the expression of CP1 (Supplemental Figure 6D). Ji et al. (2013) reported that mutation in DTD (an analog of EAT1, bHLH141) in rice results in severe male sterility. The eat1 or dtd mutant undergoes a normal meiosis process but ms142 cannot proceed beyond meiotic cell division (Supplemental Figure 2). Similar to udt1 (Jung et al., 2005), ms142 meiocytes did not develop to microspores, while mutants of TDR1 in rice and its homolog AMS in Arabidopsis showed thickened tapetum and microspore degeneration after release from tetrads (Sorensen et al., 2003;Li et al., 2006a).
Our transmission electron microscopy (TEM) examination demonstrated abnormal endoplasmic reticulum (ER) and apoptosis in MMCs of ms142 mutant at the early meiosis stage ( Figure 2N). This implies that bHLH142 might play an extra role in meiocyte development. So far, very little has been reported on ER apoptosis in relation to plant pollen development. The ER functions in the synthesis and folding of proteins. Correctly folded proteins exit the ER and transport to the Golgi and other cellular compartments. Misfolded proteins may accumulate in the ER and cause stress to the cells. In response, cells may activate the unfolded protein response to maintain homeostasis of proteins in the ER. If the unfolded protein response is not sufficient to recover normal ER function, cells die by apoptosis (Rasheva and Domingos, 2009).
Our ISH analysis revealed that as well as being expressed in the tapetal layer, bHLH142 is also expressed in meiocytes as well as in the vascular bundle ( Figure 5). These results suggest that bHLH142 might also play an additional role in nutrient acquisition during microspore development. Our qRT-PCR data demonstrated that expression of CYP703A3, CYP704B2, MS2, and C6 is significantly downregulated in ms142 anthers ( Figure  6), confirming the prediction of a coexpression network by Aya et al. (2011). Therefore, our data support the notion that (A) Y2H assays. Constructs expressing the full-length bHLH142 were cloned into the prey vector pGADT7 (AD) and truncated forms of TDR and EAT1 were introduced into the bait vector pGBKT7 (BD), respectively. Yeast strains were tested for growth in selective medium lacking Leu, Trp, and His and supplemented with 3-amino-1,2,4 triazole and X-Gal. (B) BiFC in rice protoplasts expressing the indicated constructs. The top panel shows bHLH142 protein interacted with TDR in the nucleus. The bottom panel shows TDR interacted with EAT1 in the nucleus. NLS-RFP was used as nuclear localized control. Bar = 10 µm. (C) Co-IP assay of HA fused TDR and bHLH142 recombinant proteins expressed in Escherichia coli, pulled down by protein G fused with bHLH142 antibody, and detected using anti-HA antibody. (D) Co-IP assay of HA fused TDR and bHLH142 recombinant proteins expressed in E. coli, pulled down by protein G fused with HA antibody, and detected using anti-bHLH142 antibody.
bHLH142 plays an additional role in the sporopollenin biosynthesis pathway.
This study uncovers bHLH142 as a bHLH TF family factor that is critical for pollen development alongside UDT1 (bLHL164), TDR1 (bLHL5), and EAT1 (bHLH141). Our data suggest that bHLH142 acts downstream of UDT1 (bHLH164) but upstream of TDR1 (bHLH5) and EAT1 (bHLH141) (Figure 6). Interestingly, all four bHLH TFs are expressed specifically in the anther and (A) Predicted amino acid patches on bHLH142 and TDR1 that form protein-protein interactions. (B) Predicted amino acid patches on EAT1 and TDR1 that form protein-protein interactions. (C) The predicted heterodimers of bHLH142/TDR1 and EAT1/TDR1. The 3D structure modeling indicated that both bHLH142 and EAT1 bind to TDR1 at the same region in the C termini and form a similar interface.
bHLH142 Regulates Pollen Development in Rice participate in sequential pollen development events, particularly in tapetal PCD. Several of the bHLH TFs act coordinately in regulating anther development, and it is thus likely that more TFs are involved in controlling the regulatory network to ensure normal pollen development.
Also, we noticed that there was lower level of expression of TDR1 in ms142 compared with the other downstream genes in the regulatory network. This may be attributed to the fact that TDR1 is also known to be regulated by another TF, GAMYB (Liu et al., 2010). In agreement with this finding, the expression of GAMYB was not altered in ms142 (Figure 6). Taken together, these results suggest that two parallel pathways may exist in the regulatory circuit leading to TDR1 during pollen development.
bHLH142 Functions Coordinately with TDR1 to Regulate the EAT1 Promoter Transcripts of TDR1 and EAT1 are both downregulated in the ms142 (Figures 6E and 6F). We hypothesize that TDR1 interacts with bHLH142 and positively regulates EAT1 expression and thus the transcriptional activities of AP25 and AP37, which encode aspartic proteases for tapetal PCD. Our promoter transient assay provides solid evidence that bHLH142 and TDR1 work coordinately in regulating the EAT1 promoter (Figure 7). We also demonstrated that additional expression of the EAT1 protein significantly reduced EAT1-Luc promoter strength from a 30fold increase down to an 18-fold increase (Figure 7), which may be attributed to the competition between bHLH142 and EAT to interact with TDR1. Presumably, more EAT1 favors the TDR1/ EAT1 interaction and might consequently reduce the interaction between bHLH142 and TDR1, therefore reducing EAT1 transcriptional activation (Figure 7). Our data suggest that bHLH142 interacts with TDR1 to coregulate EAT1 expression.
Our molecular studies provide solid in vivo (Y2H and BiFC) and in vitro (co-IP) evidence that bHLH142 and TDR1 proteins interact (Figure 8). The co-IP provides convincing evidence that these two proteins physically interact in vitro. Subcellular localization also demonstrates that bHLH142 protein is localized in the nucleus (Supplemental Figure 4; Figure 8B) and its protein is not self-activated (Supplemental Figure 7A). Since we also found self activation of full-length TDR1 and EAT1 in our Y2H experiment (Supplemental Figure 7), N-terminal truncated forms of TDR Δaa(1-344) and EAT1 Δaa(1-254) were used in our experiments to reduce self-activation. These two N-terminal truncated protein forms did not exhibit self-activation in yeast cells (Supplemental Figure 7A). Therefore, we are confident that bHLH142 interacts with TDR1 by using these truncated proteins to eliminate the bias (Figure 8). Our data indicate that bHLH142 interacts with TDR1 in the C-terminal (Figure 8) and support the conclusion of Ji et al. (2013) that DTD/EAT1 (bHLH141) interacts with TDR1 in the C-terminal region. In other words, both bHLH142 and EAT1 (bHLH141) can interact with TDR1 in the C terminus of TDR1. This finding also supports the result of our EAT1 promoter assay, which showed that additional EAT1 protein reduced EAT1 promoter activity, presumably due to the competition between bHLH142 and EAT1 proteins in the C terminus of TDR1.
From the 3D modeling, it is clear that bHLH142, EAT1, and TDR1 all have electrical and hydrophobic interactive surface patches in the C-terminal region for protein-protein interaction (Figure 9). Both bHLH142 and EAT1 can form a heterodimer with TDR1 with a favorable match and form a similar interface. We also predicted a favorable DNA binding pocket on the bHLH142/ TDR1 heterodimer to bind to the specific E-box sequences (CAATTG) on the promoter of EAT1 to trigger its transcription ( Figure 10). Presumably, once the dimer binds to the E-box, the nucleotide base pairs on both ends of the DNA molecule begin to dissociate and loosen the DNA helix structure for transcriptional activation. These modeling results are consistent with our Y2H data in this study that suggest that bHLH142 interacts with the C-terminal of TDR1 ( Figure 8) and our transient promoter assay that demonstrated that simultaneous expression of bHLH142 and TDR1 is required for the expression of EAT1 (Figure 7). The predicted dimerization between EAT1 and TDR1 (Figure 9) is also consistent with the data of Niu et al. (2013) and Ji et al. (2013), who showed from Y2H analysis that TDR1 can interact with EAT1. Our modeling suggests that EAT1 competes well with bHLH142, binding to TDR1 in a similar manner with a comparable binding (A) The TDR1/bHLH142 heterodimer binds to an E-box (CAATTG) on the EAT1 promoter, with the E-box fitting to the pocket center at the interface of the heterodimer. (B) The 6-bp E-box (CAATTG) binds to TDR1 at 67 Leu-75 Leu, 231 Val -245 His, and 352 Asp -363 Leu and to bHLH142 at 8 Leu-26 Ala and 115 Ala-123 Gly. free energy. Therefore, we propose that the interaction between bHLH142 and TDR1 promotes the expression of EAT1, which in turn interacts with TDR1 and serves as a feedback inhibitor of EAT1 transcription. In agreement with this proposal, simultaneous expression of both bHLH142 and TDR1 largely enhanced EAT1 expression, while additional expression of EAT1 along with bHLH142 and TDR1 competitively reduced the expression of EAT1 (Figure 7). Our genetic study clearly showed that the bHLH142 knockout mutant has significantly lower EAT1 mRNA as shown by qRT-PCR analysis ( Figure 6F). Taken together, our modeling results are consistent with our experimental data in supporting the pivotal role of bHLH142 in EAT1 expression and male sterility in rice. Whether EAT1/TDR1 heterodimer has other biological functions in gene regulation requires further investigation.

The Regulatory Cascade of Tapetal PCD in Rice
Based on this study and previous studies, an updated regulatory network for rice pollen development is presented in Figure 11.
Previous work with various rice MS mutants suggests that UDT1 and GAMYB may positively regulate the transcription of TDR1 (Liu et al., 2010), and TDR1 controls the transcription of C6 and CP1 (Li et al., 2006a). Recently, Niu et al. (2013) presented evidence that TDR1 interacts with EAT1 and that EAT1 directly regulates the expression of two aspartic proteases for initiation of tapetal PCD. In this study, we demonstrate that bHLH142 acts downstream of UDT1, but upstream of TDR1 and EAT1, and then bHLH142 interacts with TDR1 to activate EAT1 transcription by binding to its promoter. In addition, we also provide evidence that EAT1 may directly regulate the expression of CP1. Therefore, we conclude that bHLH142 is essential for tapetal PCD and pollen development in rice.

Application of Genetic Male Sterility in Hybrid Crop Production
Crops produced from F1 hybrid seeds offer significant benefits in terms of yield improvement, agronomic performance, and consistency of end-use quality. This is due to the hybrid vigor The model depicts the molecular function of bHLH142 in rice anther development, relative to other known key regulators. bHLH142 acts downstream of UDT1 but upstream of TDR1 and by interacting with TDR1 to form a heterodimer (Figure 9). Both bHLH142 and TDR1 coordinately regulate the expression of EAT1 by binding its promoter (Figure 8). In turn, EAT1 interacts with TDR1 (Ji et al., 2013;Niu et al., 2013) and forms a heterodimer in a similar manner (Figure 9). Thus, EAT1 competes with bHLH142 for the similar protein binding sites on TDR1 with a comparable binding free energy ( Figure 9); therefore, feedback inhibits bHLH142/TDR1 interaction and EAT1 expression (Figure 7). Previous studies suggest that TDR1 acts downstream of UDT1 and GAMYB and that UDT1 and GAMYB work in parallel to regulate early stage anther development (Liu et al., 2010). TDR1 directly regulates the expression of C6 and CP1 that encode cysteine protease and protease inhibitor, respectively (Li et al., 2006a), while EAT1 regulates the expression of two aspartic proteases (AP37 and AP25) and promotes tapetal PCD during anther development (Niu et al., 2013). This study also provides evidence that EAT1 may directly regulate the expression of CP1. Evidence from previous studies is indicated by white arrows, while data demonstrated in this study are indicated by red arrows. The background is TUNEL signal in anther of TNG67 at the young microspore stage. The red signal is propidium iodide staining, and yellow fluorescence is the merged signal from TUNEL (green) and propidium iodide.
bHLH142 Regulates Pollen Development in Rice generated by combining carefully selected parent lines. Hybrid crops are responsible for the dramatic increase in global crop yields in recent decades, and male-sterile traits have played a significant role in this advancement. CMS, a maternally inherited trait, has been widely exploited for hybrid crop breeding (e.g., maize and rice) due to the convenience of controlling sterility expression by manipulating the gene-cytoplasm combinations in any selected genotype. Most importantly, it eliminates the need for emasculation in cross-pollinated species, thus encouraging cross breeding, producing pure hybrid seeds under natural conditions. However, commercial seed production must be simple and inexpensive, and the requirement for a maintainer line to produce the seed stocks of the CMS line increases the production cost for this three-line hybrid system. GMS, controlled by nuclear genes, offers an alternative hybrid seed production system. For the two-line hybrid system, it is beneficial to use photoperiod-or temperature-inducible male sterility (PGMS or TGMS) mutants to maintain seed stocks for hybrid seed production. In China, PA64S is currently the most widely used maternal line in two-line hybrid rice breeding. It is crossed with paternal line 93-11 to generate superhybrid rice, LYP9 (Luo et al., 1992). PA64S, derived from a spontaneous PGMS japonica mutant NK58S (long day, >13.5 h; Shi, 1985), is also a TGMS indica rice, whose male sterility is promoted by temperatures greater than 23.5°C, but recovers its fertility at temperatures between 21 and 23°C. Recent mapping analyses demonstrate that the PGMS/TGMS in these lines is regulated by a novel small RNA (Zhou et al., 2012). In another rice genic malesterile mutant discovered recently, Carbon Starved Anther (CSA), a mutation of the R2R3 MYB transcription regulator results in a defect in pollen development (Zhang et al., 2010b). Further study of this mutant showed that csa is photoperiodsensitive, exhibiting male sterility under short-day conditions but male fertility under long-day conditions . The molecular basis of this sensitivity to daylength remains undetermined.
Transgenic male sterility has been generated using a number of transgenes, but its application in commercial production of hybrid seeds is limited due to the lack of an efficient and economical means to maintain the male sterile lines or the lack of suitable restorers (Li et al., 2007). Recently, a reversible malesterile system has been demonstrated in transgenic Arabidopsis plants by manipulating a R2R3 MYB domain protein (MYB103) (Li et al., 2007). Blocking the function of MYB103 using an insertion mutant or an MYB103EAR chimeric repressor construct under the control of the MYB103 promoter resulted in complete male sterility without seed setting (Li et al., 2007). A restorer containing the MYB103 gene driven by a stronger anther-specific promoter was introduced into pollen donor plants and crossed into the male-sterile transgenic plants for the repressor restoring the male fertility of F1 plants. The chimeric repressor and the restorer constitute a reversible male-sterile system for hybrid seed production. The successful application of this system for largescale hybrid seed production depends on whether the female parent lines can be multiplied efficiently and economically. Alternatively, a promoter inducible by chemicals or other factors (e.g., photoperiod or temperature) can be directly used to regulate the expression of a GMS gene (e.g., bHLH142) and control pollen development in transgenic plants, eliminating the costly need to maintain male-sterile lines.

Plant Materials and Growth Conditions
Seeds of the ms142 mutant were obtained from the TRIM library (http:// trim.sinica.edu.tw/). Seedlings of ms142 mutant and its wild type (TNG67) were raised in half-strength Kimura solution for 3 weeks and then transplanted into soil in the Academia Sinica-Biotechnology Center in Southern Taiwan screen house for genetically modified organisms, located in Tainan, Taiwan.

Anther Anatomy
Spikelets and anthers of the wild type and ms142 mutant were sampled at various stages of development and fixed overnight in phosphate buffer (pH 7.0) that contained 4% paraformaldehyde and 2.5% glutaraldehyde. They were then rinsed with the same buffer and postfixed for 30 min in phosphate buffer, pH 7.0, containing 1% osmium tetroxide. After dehydration, the specimens were embedded in Spurr's Resin (EMS). The processor, KOS Rapid Microwave Labstation, was chosen for postfixation, dehydration, resin infiltration, and embedding. For TEM, ultrathin sections (90 to 100 nm thick) collected on coated copper grids were stained with 6% uranyl acetated and 0.4% lead citrate and examined using transmission electron microscope.

Meiotic Analysis
For meiotic studies, young spikelets of the wild type and ms142 at 2-to 4mm length stages were fixed with ethanol:acetic acid (3:1, v/v) overnight, then transferred to 70% ethyl alcohol and stored under refrigeration. Meiotic chromosome spreads were prepared following the protocol described previously (Chang et al., 2009b) with some modifications. In brief, anthers were dissected gently with forceps and needles and transferred to a glass slide together with a drop of 1% acetocarmine and squashed gently with a needle to release MMCs; and the anther wall debris was carefully removed before a cover slip was added. Finally, the preparation was gently squashed by vertical press and analyzed under microscopy (Zeiss Axio Scope A1). Photographs were taken using Axio-Cam HRc camera (Zeiss).

TUNEL Assay
PCD is characterized by cellular condensation, mitochondria and cytoskeleton degeneration, nuclear condensation, and internucleosomal cleavage of chromosomal DNA (Phan et al., 2011). To investigate the nature of the tapetal breakdown in ms142, TUNEL assay was performed using the DeadEnd Fluorometric TUNEL system (Promega). This assay detects in situ DNA cleavage, a hallmark feature of apoptosis-like PCD, by enzymatically incorporating fluorescein-12-dUTP into the 39-OH ends of fragmented DNA. Stage of anther development was based primarily on spikelet size and developmental stages.

Total RNA Isolation and PCR
Total RNA was isolated from rice (Oryza sativa) tissues using MaestroZol RNA PLUS (Invitrogen) as described by the supplier. Various rice organs at different developmental stages were harvested for RNA isolation: root, shoot, flag leaf, internode, and panicles of 0.5, 1, 5, 9, 20 cm length, spikelets at 1 d before anthesis, lemma, palea, anthers, ovary, and seeds at 5 d after pollination (S1), 15 d after pollination (S3), 25 d after pollination (S5), and calli. The stages of anthers were classified into the following categories according to spikelet length: MMC with spikelet length of ;2 mm, meiosis (4 mm), young microspore (YM, 6 mm), vacuolated pollen (VP, 8 mm), mitosis pollen (MP, 8 mm with light-green lemma), and mature pollen at 1 d before anthesis. Total RNA was treated with DNase (Promega), and 1 µg RNA was used to synthesize the oligo(dT) primed first-strand cDNA using the M-MLV reverse transcriptase cDNA synthesis kit (Promega). One microliter of the reverse transcription products was used as a template in the PCR reactions. Ubiquitin-like 5 and 18srRNA were used as normalizer control. Each sample had three biological repeats.

qRT-PCR Analysis
Fifteen microliters of RT-PCR reaction contained 4 mL of 1/4 diluted cDNA, 3 µM of primers, and 7.5 mL of 2X KAPA SYBR FAST master mix (KAPA Biosystems, USA). Real-Time PCR was performed using a CFX96 Real-Time PCR detection system (Bio-Rad, USA). Quantification analysis was performed using CFX Manager Software (Bio-Rad, USA). Primers used for qPCR are listed in Supplemental Table 4.

In Situ Hybridization
Spikelets of TNG67 and ms142 at various developmental stages were collected and fixed, dehydrated, embedded, sliced, and performed hybridization as previously described (Lin et al., 2014). For preparation of digoxigenin-labeled RNA probes, we cloned the gene-specific region of bHLH142 using primers listed in Supplemental Table 4.

Subcellular Localization of bHLH142
For subcellular localization of bHLH142, the coding sequences of the gene were subcloned into p2FGW7 (Invitrogen) to generate bHLH142-GFP fusion genes driven by the cauliflower mosaic virus 35S (CaMV35S) promoter. Rice protoplasts were isolated and transformed using the polyethylene glycol method following procedures described previously (Bart et al., 2006). After incubation at room temperature for 16 h in light, protoplasts were observed with a Zeiss LSM 780 confocal laser scanning microscope.

Phylogenetic Analysis of the bHLH142 Subfamily
The bHLH142 protein sequence was used to search for the closest homologs from the plant species using BLASTP programs. Multiple sequence alignment of full-length protein sequences was performed using ClustalW online (http://www.ch.embnet.org/software/ClustalW. html), and the alignment was used to perform neighbor-joining analysis using Mega 5.05 (Tamura et al., 2011). The numbers at the nodes represent percentage bootstrap values based on 1000 replications. The length of the branches is proportional to the expected numbers of amino acid substitutions per site. Gene accession numbers of the sequences used to generate the phylogenetic tree and the alignment can be found as Supplemental Data Set 1.

Rice Protoplast Transient Expression and Reporter Gene Activity Assay
Transient promoter assay was performed as previously described  with some modifications. In brief, protoplasts were isolated from leaf tissue of 10-d-old rice seedlings. The reporter plasmid contained the CaMV35S minimal promoter and the EAT1 promoter (2031 bp) fused to the firefly luciferase gene (Luc). In the effector plasmids, bHLH142, TDR1, and EAT1 genes were under the control of the CaMV35S promoter. The pBI221 vector contained a CaMV35S promoter for driving the expression of GUS as an internal control.

Y2H Assay
The Matchmaker GAL4 two-hybrid system (Clontech) was used for Y2H assays. Since both full-length EAT1 and TDR1 proteins were reported to have self-activation (Ji et al., 2013), we made a truncated EAT1 (EAT1 D , amino acids 1 to 254) and a truncated TDR1 (TDR D , amino acids 1 to 344) to reduce self-activation. The full-length cDNA of bHLH142 was cloned into pGAD-T7 (Clontech), and full-length bHLH142, EAT1, TDR, EAT1 D , and TDR D were cloned into pGBK-T7 (Clontech), respectively. The pairs of constructs to be tested were cotransformed into AH109 yeast cells and selected on plates containing Leu (for pGADT7 plasmid) and Trp (for pGBKT7 plasmid) dropout medium for 3 to 4 d at 30°C. Transformants were tested for specific protein interactions by growing on SD/-Leu/-Trp/-His plates with 30 mM 3-amino-1,2,4 triazole and tested after X-a-Gal induction to confirm positive interaction. This system provides a transcriptional assay for detecting and confirming protein interactions in vivo in yeast.

BiFC Assay
BiFC assay allows visualization of protein-protein interactions in living cells and the direct detection of the protein complexes in subcellular compartments, providing insights into their functions. Full-length cDNAs of bHLH142, TDR1, and EAT1 were independently introduced into pJET1.2 (Thermo Scientific). The sequence for the N-terminal amino acid residues 1 to 174 of YFP was then in-frame fused to the sequence of the C-terminal region of the tested proteins, while the sequence of the C-terminal amino acid residues 175 to 239 of YFP was in-frame fused to the sequence of the N-terminal end of the proteins. Next, the tested genes were introduced into pSAT5-DEST_CYN1 and pSAT4(A)-DEST_NYN1. Ballistic bombardment-mediated transient transformation in rice protoplasts was performed following a previously published protocol (Hsu et al., 2011). Florescence images were photographed on a LSM 780 Plus ELYRA S.1 confocal microscope with a Plan-Apochromat 403/1.4 oil objective lens (Zeiss).

Co-IP Assay
Recombinant proteins of bHLH142 and TDR1 fused with HA tag were expressed in bacteria harboring pET-53-DEST (HIS-tag), and cell extracts after lysis were centrifuged at 12,000 rpm for 15 min, suspended in binding buffer (20 mM Tris-HCl, pH 7.9, and 500 mM NaCl), and sonicated on ice for 30 s using an ultrasonic homogenizer (Misonix XL Sonicator Ultrasonic Cell Processor). The supernatants were purified using Ni 2+ resin. For immunoprecipitation, extracts were precleared by 30 min incubation with 20 mL of PureProteome Protein G Magnetic Beads (Millipore) at 4°C with rotation. The antibodies (anti-bHLH142 or anti-HA) were then added to the precleared extracts. After incubation for 4 h at 4°C, 40 mL of PureProteome Protein G Magnetic Beads was added, and the extracts were further incubated for 10 min at room temperature with rotation. After extensive washing, bound proteins were analyzed by immunoblotting. Rabbit antiserum against rice bHLH142 was produced using a synthetic peptide (CSPTPRSGGGRKRSR) as antigen (GenScript).

Prediction of 3D Protein Structure and Interaction by Molecular Dynamics Simulation
3D protein structures were constructed by MD simulation using ab initio modeling methods. MD simulation was performed with the program NAMD (Nelson et al., 1996) using parameters adopted from the CHARMM force field (Brooks et al., 1983). The full-length protein sequences of EAT1 (NP_001053749.1), bHLH142 (NP_001042795.1), and TDR1 bHLH142 Regulates Pollen Development in Rice (NP_001045710.1) were obtained from the National Center for Biotechnology Information database. The models were minimized by removing unfavorable contacts, brought to 310K by velocity rescaling, and equilibrated for 1 ns. Before any MD trajectory was run, 40 ps of energy minimization was performed to relax the conformational and structural tensions. The minimum structure was the starting point for MD simulation. For this purpose, the protein molecule was embedded into a water simulation box and a cutoff distance of 12 Å was employed for the nonbonded and electrostatic interactions. The heating process was performed from 0 to 310K through Langevin damping with a coefficient of 10 ps 21 . A time step of 2 fs was employed for rescaling the temperature. After 20 ps of heating to 310K, equilibration trajectories of 2 µs were recorded, which provided the data for the structural and thermodynamic evaluations. The equations of motion were integrated with the Shake algorithm with a time step of 1 fs. The modeling quality of 3D models was evaluated by MQAP (Fischer, 2006) and the CASP7 experiment methods of QA (Cozzetto et al., 2007). Figures displaying atomistic pictures of molecules were generated using UCSF Chimera (Pettersen et al., 2004). The KD hydrophobicity of amino acid patches on protein surface was labeled as described previously (Kyte and Doolittle, 1982).

RNAi-Mediated Gene Silencing of bHLH142
To generate an RNAi construct for suppressing the expression of bHLH142, a 149-bp fragment from the 59 UTR of bHLH142 was amplified by PCR with specific primers (Supplemental Table 4) and cloned into pENTR (Invitrogen) to yield an entry vector pPZP200 hph-Ubi-bHLH142 RNAi-NOS (12,483 bp). The RNAi construct was transformed into wildtype (TNG67) rice calli via an Agrobacterium tumefaciens-mediated transformation system (Chan et al., 1993). Transgenic plants were regenerated from transformed calli by selection on hygromycin-containing medium.

Supplemental Data
The following materials are available in the online version of this article.   Figure 5.