The tapetum (T) is formed through periclinal cell division of somatic cells and corresponds to the innermost of four characteristic anther layers, which also include the epidermis (Ep), the endothecium (En), and the middle layer (ML). The tapetum surrounds the growing microspores in the anther locule and plays a key role in male gametogenesis by providing nutrients and other necessary ingredients that ensure proper pollen and anther development. Thus, any defect during the differentiation process of the tapetum results in male sterility (Ariizumi and Toriyama, 2011). The differentiation of the tapetum and meiosis occur simultaneously in rice. A number of studies have previously specified the role played by different genes, including transcription factors (TFs), during tapetum differentiation. The rice undeveloped tapetum1 (UDT1) gene encodes for a bHLH TF that is involved in triggering the tapetum differentiation by terminating the periclinal cell division of somatic cells. udt1 mutants exhibited an undeveloped and vacuolated tapetum and did not allow for the development of meiocytes into microspores (Jung et al., 2005). Similarly, rice mutants of TDF1 (OsTDF1), which is an R2R3 MYB TF, displayed a defective and vacuolated tapetum, suggesting that this gene also plays a crucial role in tapetum development and male fertility (Cai et al., 2015). Furthermore, GAMYB encodes a transcription factor associated with the Gibberellin (GA) pathway and regulates anther development. Mutations in GAMYB are known to cause an expanded and undifferentiated tapetum and lead to an abnormal meiosis (Aya et al., 2009; Liu et al., 2010). Additionally, two bHLH transcription factors, TAPETUM DEGRADATION RETARDATION (TDR) and ETERNAL TAPETUM1 (EAT1), are also essential for tapetum development (Li et al., 2006; Niu N. et al., 2013). Finally, the rice MYB80 (OsMYB80) gene regulates anther development by targeting multiple pollen development pathways, as myb80 mutants displayed premature PCD, an absence of Ubisch bodies, and pollen exine that resulted in no pollen production (Pan et al., 2020).

Meiosis is a crucial step for the division of PMCs, which is necessary to ensure equal chromosomal segregation to daughter cells. While the meiotic process can be crucial to determine male fertility, its role in anther and pollen wall development largely remains undetermined. Recently, a few genetic and cytological studies have evaluated the role of meiosis-related genes on pollen development. The rice gene DEFECTIVE TAPETUM AND MEIOCYTES 1 (DTM1) encodes an ER membrane-localized protein and is crucial for the differentiation of the tapetum and meiotic progression during early anther development. The disruption of DTM1 arrests the meiotic process during prophase 1 and delays ER degradation in the tapetum, which eventually causes complete male sterility (Yi et al., 2012). Furthermore, mutations in a meiosis-specific gene, FINGL1, cause an abnormal meiosis and lead to severe defects in the anther and pollen walls due to an excessive deposition of sporopollenin and anther cutin (Zhang et al., 2017). Finally, the gene OsMFS1 encodes for a meiotic coiled-coil protein that is essential for the formation of double-strand breaks, which plays an indispensable role in meiosis and anther development. Accordingly, mfs1 mutants produce abortive pollen with reduced sporopollenin deposition, an impaired tectum, and columella formation (abnormal exine) (Lu et al., 2020).

The transcriptional regulatory network can be divided into three pathways based on the mode of activation of target genes by the TFs. Specifically, these pathways include, first, direct activation: In this pathway, the TFs directly regulate transcription by directly binding to the promoter of target genes. For example, the GAMYB encoding transcription factor is essential for the Gibberellin (GA) pathway and is involved in the direct regulation of expression of the genes OsC6, OsCYP703A3, and OsKAR (Aya et al., 2009; Zhang H. et al., 2010). Similarly, the TFs Undeveloped Tapetum 1 (UDT1) and Persistent Tapetal Cell 1 (PTC1) directly bind to the promoter of a GDSL lipase-encoding gene rice male sterile2 OsRMS2/OsGELP34 and regulate its expression (Zhao et al., 2020). The genes TDR and EAT1 directly regulate the expression of OsCP1 and OsAP25, respectively (Li et al., 2006; Niu B.X. et al., 2013). Second is the indirect regulation: In this pathway, one TF regulates other TF(s), which, in turn, activate the expression of downstream target gene(s). The TDR Interacting Protein 2 (TIP2) and TDR Interacting Protein 3 (TIP3) genes follow the pathway of indirect regulation of target genes. Moreover, TIP2 and TIP3 regulate the expression of the Tapetum Degeneration Retardation (TDR) gene (Fu et al., 2014; Ko et al., 2014; Yang et al., 2019b), which, in turn, regulates the expression of other target genes, including OsC6 and OsCYP703A3, by directly binding to their promoter. Similarly, the gene TGA10 interacts with both TIP2 and TDR and affects the expression of AP25 and MTR1 genes (Chen et al., 2018). Third is the coordinated regulation: Different TFs are involved in a cascade reaction by interacting with each other and synergistically regulate the expression of common target genes. For example, three bHLH proteins, specifically TDR INTERACTING PROTEIN 2 (TIP2), Tapetum Degeneration Retardation (TDR), and ETERNAL TAPETUM 1 (EAT1) interact with each other to form a regulatory cascade and share common target genes, such as AP25 and AP37 (Fu et al., 2014; Ko et al., 2014; Niu B.X. et al., 2013).

Nongken 58S was the first reported photoperiod-sensitive male sterile (PGMS) line in rice, which was caused by a G-to-C single-nucleotide polymorphism (SNP) point mutation that alters the secondary structure of a lncRNA, termed long-day-specific male-fertility-associated RNA (LDMAR) (Ding et al., 2012a). Transcriptome and deep sequencing analysis of Nongken 58S detected three small RNAs within LDMAR (Ding et al., 2012a). However, a follow-up study found that the DNA methylation patterns observed in the promoter region of LDMAR is most likely directed by a 21-nt small RNA (Psi-LDMAR), which led to reduced transcript production and thus to male sterility under long-day conditions (Ding et al., 2012b) (Figure 2A). Later, the PMS3 locus was isolated in the TGMS line Peiai64S (which was bred by crossing NK85S and Peiai64) and named P/TMS12-1 (Zhou et al., 2012). PMS3 causes TGMS and PGMS in the Peiai64S and NK85S lines, respectively. The P/TGMS (Peiai64S) lines exhibit a point mutation in a 21-nt small RNA (osa-smR5864m), which may lead to loss-of-function and subsequent failure to target different genes in the indica and japonica backgrounds, and produces PGMS and TGMS, respectively (Zhou et al., 2012). Accordingly, these studies support the involvement of DNA methylation or RNA-dependent DNA methylation (RdDM) in the regulation of PTGMS.

The pms1 locus, a 21-PHAS gene, encodes a long non-coding RNA (lncRNA) PMS1T in the PGMS rice line Nongken 58S that is specifically expressed in young panicles. Upon production, this lncRNA undergoes cleavage after being targeted by miR2118 and produces an abundance of 21-nt phased small-interfering RNAs (phasiRNAs) under long-day conditions (Figure 2B). Furthermore, a single point mutation (G to A) in the lncRNA close to the miR2118 recognition site results in fertility–sterility transition most likely by the differential accumulation of phasiRNAs (Fan et al., 2016). Under short-day conditions, this mutation does not affect phasiRNA production and instead produces a fertile phenotype that is similar to the fertile line Nongken 58N (Fan et al., 2016).

Increasing evidence suggests that the previously regarded non-coding DNA (dark matter DNA) in the genome, no longer seems non-functional. For example, the microRNA2118 (miR2118) is conserved in plants and involved in the production of phasiRNA. Knockout miR2118 mutants display severe male and female sterility problems in rice with marked developmental and morphological abnormalities in the somatic cell layers of the anther wall, especially under short-day conditions. In contrast, the severity of these phenotypes is lower, and plants are fertile under long-day conditions (Figure 2C). The miR2118 mutants show an enrichment of U rich 21-nt phasiRNAs in anthers, suggesting that the loss of miR2118 function impairs the production of 21-nt phasiRNAs and thus causes sterility. Furthermore, proteome analysis proposed that the argonaute proteins OsAGO1b/OsAGO1d may be involved in an miR2118-dependent 21-nt phasiRNAs biogenesis (Araki et al., 2020). These results suggest that miR2118 presents a novel mechanism of reproductive development as an environmental response because fertility is partially dependent on the photoperiod but is independent of other photoperiodic pathways.

Thermosensitive genic male-sterile (TGMS) lines constitute a very important source for hybrid rice breeding. TMS5 is a major sterility–fertility transition gene that was well investigated in a number of studies focusing on fertility transition under different temperature conditions. TMS5 encodes a conserved RNase Z protein (RNase Z^S1), which regulates the levels of accumulation of the ubiquitin-60S ribosomal protein L40 (Ub_L40) messenger RNA (mRNA) (Figure 3A). The RNase Z^S1 cleaves the ubiquitin fusion ribosomal protein L40 (Ub_L40) mRNAs to maintain fertility. The tms5 mutants are unable to produce RNase Z^S1, which leads to an overaccumulation of unprocessed Ub_L40 mRNAs under high temperature conditions, and results in the production of defective pollen and male sterility (Zhou et al., 2014).

The sterility–fertility transition of TGMS line (Zheda13S) derived from Peiai64S and carrying the tms5 gene was investigated via RNA-seq analysis, which detected differentially expressed genes (DEGs) related to different cellular functions, such as protein folding, protein binding, transcription regulation, transcription factor activity, and metabolism. Furthermore, topological analysis detected Ub_L40s with decreased expression, DNA-directed RNA polymerase subunit, kinases, and heat shock proteins (HSPs), all of which can physically interact with each other. In summary, the sterility of Zheda13S under restrictive (high) temperature is linked to DNA-directed RNA polymerase subunits, kinases, HSPs, and other DEGs, along with a higher accumulation of Ub_L40s mRNA (Li et al., 2020). It can thus be concluded that there exists a number of other factors that collectively control the expression of the tms5 gene.

Leucine-rich repeats receptor-like kinases are widely associated with abiotic stress responses in plants, even though their contribution to male sterility has not yet been properly studied. The rice gene Thermosensitive Genic Male Sterile 10 (TMS10) and its homolog TMS10-Like (TMS10L) encode a leucine-rich repeats receptor-like kinase (LRR-RLK) and perform redundant functions in controlling tapetal development. Mutants of these genes produce sterile phenotypes at high temperatures and fertile phenotypes at low temperature (Yu et al., 2017). Moreover, the double mutants of tms10 and tms10l exhibit complete male sterility at both high and low temperatures. These results suggest that TMS10 is an excellent genetic resource for the production of hybrid rice varieties.

The pollen coat is the outermost layer of pollen grains, protecting the released pollen from desiccation and other environmental hazards in order to ensure a successful pollination (Pacini and Hesse, 2005). However, pollen coat composition and function is sometimes compromised by environmental fluctuations. OsOSC12/OsPTS1 encodes a triterpene synthase, and defective OsOSC12 mutants exhibit a reduced level of C16 and C18 fatty acids in the pollen coat and produce conditional male sterility under low relative humidity levels (Xue et al., 2018). Gas chromatography–mass spectrometry (GC-MS) analysis of pollen coat extracts revealed significantly decreased levels of palmitic (C16) and linolenic (C18) acids, despite no changes were observed in the levels of stearic acid (C18). The differential presence of these three important fatty acids caused pollen desiccation (Xue et al., 2018). Similarly, the rice gene HUMIDITY SENSITIVE GENIC MALE STERILITY 1 (OsHMS1) encodes a b-ketoacyl-CoA synthase and regulates the biosynthesis of VLCFAs. OsHMS1 mutants produce male sterility in consensus with decreased VLCFAs levels (C26, C28) under low humidity conditions (Chen et al., 2020). Finally, gas chromatography–mass spectrometry detected metabolic dysfunctions and pollen desiccation due to deficiency of VLCFAs, which, as mentioned above, are a major constituent of the pollen wall (Chen et al., 2020). These genes seemingly regulate a very new type of environment-sensitive GMS known as humidity-sensitive GMS (HGMS). Although these genes are involved in lipid metabolism, their expression is vulnerable to environmental changes. These genes could be the breeding source in the regions with high humidity but may not be the subject of hybrid breeding for wider application.

There are number of transcription factors mediating the regulatory pathways of target genes (Table 2). The transcription factor GATA10, which was identified from a Wuxiang S (WXS) line derived from a tms5 mutant allele, encodes a zinc finger protein and directly activates the expression of Ub_L40 mRNA (Jin et al., 2019). The knockout mutant plants of GATA10 produce low Ub_L40 expression and tend toward fertility, suggesting that the gene is involved in the modulation of Ub_L40 for fertility conversion. GATA10 acts as a mediator for the interaction between Ub_L40s and many other TF modules to regulate TGMS traits. Another nuclear localized bHLH transcription factor, bHLH138, was identified in a 93-11 mutant line carrying a mutation in the second exon of the tms5 gene. This mutant line was unable to process Ub_L40 mRNA and exhibited typical TGMS traits. The bHLH138 forms a basic helix–loop–helix structure, binds the core region of tms5 promoter sequences, and activates its expression via the acidic amino-acid-rich domain (Wen et al., 2019). It is thus possible to conclude that the expression of GATA10 and bHLH138 can channel the expression of tms5 and the accumulation of Ub_L40 under fertility transition conditions. Since GATA10 and bHLH138 regulate the expression of tms5, their expression can be altered to regulate tms5 expression to guide male fertility.

The Carbon Starved Anther (CSA) belongs to the R2R3 MYB transcription family and is required for a different type of PGMS trait. Specifically, the CSA regulates the expression of OsMST8, a gene that encodes a monosaccharide transporter and is involved in the transportation of sugars from vegetative parts to the tapetum of developing anthers. The OsMST8 is preferentially expressed in tapetal cells and sugar-transporting vascular tissues. Mutations in the OsMST8 gene in CSA lines cause complete male sterility in both japonica and indica backgrounds under short day conditions, whereas fertility under long-day conditions display a reverse PGMS (rPGMS) trait (Zhang D. et al., 2010; Zhang et al., 2013).

The tms9-1 gene causes TGMS-like traits in HengnongS-1 and is reported as a candidate locus of PERSISTANT TAPETAL CELL 1 (PTC1) and MALE STERILITY 1 (OsMS1) (Li et al., 2011; Qi et al., 2014). The PTC1 encodes a PHD finger protein that is transiently expressed in the tapetum, and its loss of function causes complete male sterility in rice (Li et al., 2011). On the other hand, MS1 also encodes PHD finger proteins and regulates tapetal PCD and the formation of the pollen wall (Yang et al., 2019a). While tms9-1 is most likely the candidate locus for OsMS1 in HengnongS-1, it is necessary to implement a more detailed genetic analysis in order to validate this conclusion (Qi et al., 2014).

The rice gene ugp1 encodes a UDP-glucose pyrophosphorylase and is required for the deposition of callose layer surrounding developing pollen. The silencing of ugp1 causes complete male sterility that is characterized by callose disruption during meiosis. Moreover, the overexpression of ugp1 in transgenic plants also show TGMS-like traits under normal temperatures because the endogenous primary mRNA of ugp1 does not undergo splicing, leading to an overaccumulation of unprocessed mRNA, which cosuppresses ugp1 (Figure 3B). The sterile status of ugp1 can be reverted to fertility at low temperature, which is mediated by the efficient splicing of an aberrant ugp1 mRNA (Chen et al., 2007). Therefore, the cosuppression and splicing of transcripts display a novel TGMS trait.

DNA methylation contributes to environmentally induce phenotypic variations by regulating gene expression through epigenetic modification (Figure 4A). DNA methylation and whole-genome sequencing analysis have been widely applied to the study of male sterility in rice. A photoperiod-thermosensitive genic male sterile (PTGMS) line (PA64) shows a higher proportion of differentially methylated regions at high temperatures and long-day conditions and thus is sterile (Chen et al., 2014). Moreover, the rice gene BIM2 (OsBIM2) encodes a transcription factor BIM2 and is involved in brassinosteroid signaling. The hypermethylation of OsBIM2 has also been suggested to be involved in the impairment of brassinosteroid signaling and ATP production during pollen development in the PTGMS line PA64 under restrictive conditions (Hu et al., 2015).

Mutant studies have supported a regulatory role for DNA methylation in the process of male reproduction in rice. The rice gene DOMAINS REARRANGED METHYLASE 2 (OsDRM2) encodes DNA methyltransferases and is required to maintain non-CG methylation (CHG and CHH). Targeted disruption of OsDRM2 severs de novo methylation and produces pleiotropic developmental defects that result in complete male sterility along with vegetative growth defects. In addition, the osdrm2 mutant shows a deficiency in the RdDM pathway and reduced levels of genome-wide DNA methylation when compared to wild-type plants and shows a higher expression in the reproductive parts by downregulating many target genes (Moritoh et al., 2012; Anderson et al., 2013).

Histones are the major protein components of the chromatin, being responsible for its structure and functional state. Posttranslational histone modifications include ubiquitination, acetylation, phosphorylation, and methylation (Kouzarides, 2007). A variety of genomic analyses have measured the epigenetic responses on several genes associated with male reproductive development in rice (Table 3).

Two histone H2B monoubiquitination E3 ligases, OsHUB1 and OsHUB2, have been attributed a regulatory role during male reproductive development in rice (Cao et al., 2015). Specifically, loss-of-function OsHUB1 and OsHUB2 mutants cannot produce H2Bub1 and show a reduction in histone3 lysine-4 dimethylation (H3K4me2). Additionally, OsHUB1 directs histone H2B monoubiquitination (H2Bub1). Genetic analysis detected a depletion of H3K4me2, some DEGs related to tapetum degradation, and binding of H3K4me2 to the promoters of these genes. These results suggested that OsHUB1 and OsHUB2 function as E3 ligases to regulate histone H2B monoubiquitination. Depletion of H2Bub1 affects the expression of several genes during anther development, impairs gene transcription via H3K4me2, and results in complete male sterility (Cao et al., 2015). The rice gene MEIOTIC F-BOX (OsMOF) is an important E3 ligase that recruits completion of meiosis I (COM1) and radiation sensitive 51C (RAD51C) genes, diverges the ubiquitination machinery, arrests meiotic progression, and causes epigenetic double-stranded breaks (DSB) and complete male sterility in knockout mutant plants (He et al., 2016). However, details on the underlying ubiquitination mechanism are yet to be determined.

Meiosis is an indispensable process during male reproductive development in rice and is regulated by epigenetic factors. The MEIOSIS ARRESTED AT LEPTOTENE1 (MEL1) gene encodes an argonaute (AGO) protein that is specifically involved in controlling germ-cell development and meiosis in rice (Nonomura et al., 2007). MEL1 is expressed during the premeiotic phase of cell division, irrespective of gender (Nonomura et al., 2007; Komiya et al., 2014). A study performing deep sequencing of small RNAs associated with MEL1 has highlighted that the gene preferentially binds to a unique group of 21-nucleotide (nt) siRNAs carrying a conserved cytosine residue at the 5′-terminus that is highly expressed in rice reproductive organs. The loss-of-function MEL1 mutants display abnormal sporophytic and germ cell development (Komiya et al., 2014), and aberrant meiotic patterns have been observed in mel1 rice mutants during male gametophyte development (Liu and Nonomura, 2016). The transition of premeiosis to meiosis in PMCs is compromised due to an elevation in H3K9me2 and a reduction in both acetylation of lysine 9 of histone 3 (H3K9) and phosphorylation of histone 3 (H3S10) (Liu and Nonomura, 2016). These well-coordinated and comprehensive epigenetic modification events modulate MEL1, which preferentially binds 21-nt phasiRNAs. The meiotic chromosome reprogramming is severely affected in the anthers of mel1 mutants, which ultimately disturbs important meiotic events, such as the presynaptic centromere association, the initiation of DNA double-strand breaks (DSBs), and the pairing of homologous chromosomal. The rice mel1 mutants are also characterized by an abnormally developed tapetum, an unusual vacuolation of the PMCs, and a deficient chromosomal condensation during early meiosis (Nonomura et al., 2007; Komiya et al., 2014; Liu and Nonomura, 2016; Zhang Y.C. et al., 2020).

The budding uninhibited by benzimidazole 1 (Bub1), a conserved Ser/Thr protein kinase, has been reported for phosphorylation of histone H2A (Kawashima et al., 2010). In rice plants, the Bub1-RELATED KINASE1 (BRK1) gene encodes a Ser/Thr kinase and is essential for the phosphorylation of H2A by engaging SHUGOSHIN1 (SGO1). Furthermore, the inner centromeric histone H3-S10 does not undergo phosphorylation at the diakinesis stage in brk1 mutants, produces erroneous kinetochore–microtubule attachment, results in an unequal segregation of chromosomes, and produces male sterile phenotypes (Wang et al., 2012). SGO1 regulates the required homeostatic tension between the homologous kinetochores specifically at metaphase I and ensures accurate chromosomal segregation at the transition between metaphase I to anaphase I (Shi et al., 2015b).

Polycomb repressive complexes 2 (PRC2) are the only known protein family that modulates chromatin structure and, consequently, regulates target gene expression through H3K27me3 in plants (Margueron and Reinberg, 2011; Schwartz and Pirrotta, 2013). A number of genes belonging to the PRC2 protein group regulate flower development and include mutations that cause defective floral progression in plants (Tariq and Paszkowski, 2004; Liu Z. et al., 2020). In the specific case of rice, the loss-of-function mutant of EMF2B (OsEMF2B) shows complete sterility through H3K27me3-mediated enrichment of the E-function genes OsMADS1 and OsMADS6 in anthers and modulates their reproductive function (Conrad et al., 2014). Furthermore, RNA-seq analysis identified numerous long intergenic non-coding RNAs (lincRNAs), many of which belong to the PRC2 class, suggesting that they are targeted (rather than being regulated) by PRC2 and produce abnormal reproductive development (Johnson et al., 2018). Histone3 lysine4 di/tri-mrthylases (H3K4me2/3) are distributed within the rice genome euchromatin, and any disturbance can lead to severe phenotypic defects by activating other genes (Cui et al., 2013; Liu et al., 2015). The rice gene SET DOMAIN GROUP 701 (OsSDG701) encodes a H3K4-specific methyltransferase, binds chromatin to enhance H3K4me3 enrichment, promotes the expression of other florigenes (Hd3a and RFT1), and impairs the development of male gametes (Liu Z. et al., 2017).

Phased secondary siRNAs (phasiRNAs) play a key role in the regulation of plant reproductive development. The biogenesis of phasiRNAs is triggered by a 22-nt microRNA that causes an AGO protein-mediated silencing of the precursor locus PHAS. These process further recruits’ enzymes that are necessary for making sliced double-stranded mRNA that is further processed by Dicer-like (DCL) proteins into 21 and 24-nt siRNAs. These small RNAs form an RNA-induced silencing complex (RISC) with the AGO protein for transcriptional (TGS) or posttranscriptional gene silencing (PTGS). The 24-nt phasiRNAs are usually involved in meiotic stages of anther development, while 21-nt phasiRNAs play an important role during the premeiotic developmental stages of the anther and drive photosensitive male sterility (Xia et al., 2019).

The rice MULTIPLE SPOROCYTES1 (OsMSP1) gene encodes a leucine-rich-repeat receptor kinase and plays an important role in the development of anthers. This role has been detected in a study looking at an Osmsp1 male sterile mutant, which detected a depletion of miR2275 that is important to trigger the production of 24-nt phasiRNAs from PHAS loci. The depletion of miR2275 resulted in a depletion of 24-nt phasiRNAs and its precursor PHAS (Fei et al., 2016). Further expression analysis through RNA-seq found three distinct AGO protein encoding genes, OsAGO1d, OsAGO2b, and OsAGO18, suggesting a functional relationship between these genes and phasiRNA in the developmental control of male reproduction (Fei et al., 2016).

The rice Dicer-Like4 (OsDCL4) gene encodes an RNase III-like enzyme that catalyzes the processing of siRNA precursors via specific biogenesis and works in association with inverted repeat transgenes and the endogenous TRANS-ACTING siRNA3 (TAS3) to produce 24-nt siRNAs and phasiRNAs to regulate male reproductive development (Liu et al., 2007; Song et al., 2012a; Zhang Y.C. et al., 2020). The rice RNA-dependent RNA polymerases6 (OsRDR6) gene is involved in the biogenesis of variety of small RNAs, and the gene largely controls the production and accumulation of small RNAs including tasiRNAs from the TAS locus, and 21 and 24-nt phased small RNAs from PHAS locus (Song et al., 2012b).

An allelic OsRDR6 mutant (rdr6-mei) causes meiotic defects by blocking the formation of DSB. Specifically, the mutant produces decreased 21-nt RNAs and an increased proportion of 24-nt RNAs that bind to AGO protein family members and downregulate the expression of genes involved in DSB formation, which ultimately affects meiotic progression and causes no pollen type complete male sterility (Figure 4B). Furthermore, genetic analysis identified numerous differentially methylated regions, supporting a DNA methylation role for OsRDR6, which is independent of RdDM (Liu C. et al., 2020). Similarly, the biochemical function of Dicer-like3b (OsDCL3b/OsDCL5) is to process small RNAs (21 and 24 nucleotide) in rice panicles (Song et al., 2012a). The knockdown of OsDCL3b affects the biogenesis of 21- and 24-nt sRNAs, along with phased small RNAs and miRNAs. In addition, the downregulation of OsDCL3b hampers the transcription of genes related to panicle development (namely, CYP704B2 and OsPKS1), which consequently reduces pollen fertility and seed setting rate (Liao et al., 2019).

The ETERNAL TAPETUM1 (EAT1) is a bHLH transcription factor that regulates postmeiotic PCD in the anther tapetum of rice plants. EAT1 promotes the transcription of 24-PHAS RNAs and also activates DICER-LIKE5 (DCL5), which is known for processing double-stranded 24-PHASs into 24-nt lengths. Moreover, EAT1 can bind to the chromatin region of two transcription factors (TIP2 and UTD1) and act as a key regulator of meiotic phasiRNA biogenesis in the anther tapetum, as well as other bHLH proteins to regulate anther development (Ono et al., 2018).

The CMS and PTGMS are the major sterility systems being extensively applied for second-generation hybrid seed production. However, the utilization of CMS resources for hybrid seed production has a low efficiency, and PTGMS lines are often affected by environmental fluctuations. Moreover, these hybrid production methods are time consuming and cumbersome, which poses restrictions to their wider applications. The next generation male sterility-mediated hybrid production technology is an excellent platform to differentiate between GMS and maintainer seeds/lines based on seed production technology (SPT) methods. The SPT maintainer line is developed by combining (1) a wild-type fertility restorer gene, (2) a pollen killer gene, and (3) a marker gene to identify transgenic and non-transgenic seeds (Figure 5). The next generation hybrid production using SPT has streamlined the hybrid production process by taking advantage of the CRISPR/Cas system.

The transgenic construct-driven non-transgenic system of SPT has been devised and successfully applied in rice. The rice No Pollen 1 (OsNP1) gene encodes a glucose–methanol–choline oxidoreductase and is specifically expressed in the tapetum and microspores. OsNP1 has been used to construct a male sterility system by coupling the OsNP1 as a fertility restoration gene with an alpha-amylase gene (zm-aa1) to debilitate transgenic pollen and a red fluorescence protein (DsRed) gene to identify transgenic seeds. This gene construct was transformed into the osnp1 mutant to develop a transgenic SPT line, which resulted in 50% non-transgenic fertile pollen and 50% transgenic sterile pollen carrying the sterility gene. All of the produced pollen grains had an np1 genotype. The non-transgenic SPT line carried a single hemizygous transgene, and its selfing produced male sterile genotype (np1/np1) and SPT fertile genotype (np1/np1+SPT-TDNA-RFP) in 1:1. The crossing of transgenic plants raised from the fertile seeds (SPT), selected based on red florescence from self-pollinated seeds of the transgenic rice line carrying the osnp1 sterile mutant (np1/np1), produced male sterile plants. Around 85% of the F_1s of these transgenic male sterile plants crossed with fertile cultivars of different backgrounds outperformed the parental generation by 10% (Chang et al., 2016a). The male sterile mutant osnp1 and the transgene locus can be easily transform into different genetic backgrounds and facilitate the breeding of male sterile lines (Chang et al., 2016a).

Although this SPT system is associated with many potential advantages, the transmission rate of the transgene through the pollen is equal to zero, and the red-florescence-based sorting of transgenic seeds minimizes the transmission of transgenes. Still, there exists a risk associated with transgene flow, whereby the utilization of this system can be restricted in regions with strict bioethics policies.

In order to circumvent these limitations, a novel SPT-based technique was established to develop a third generation hybrid (Song et al., 2020). The rice male fertility gene CYP703A3 encodes a cytochrome P450 hydroxylase and is critical for pollen development. Furthermore, OrfH79 is a well-characterized CMS gene in the Honglian CMS (HL-CMS) system in rice (Peng et al., 2010; Wang et al., 2013). The aforementioned strategy used CYP703A3 as a fertility restorer, ofH79 as a pollen killer, and DsRed2 as the selective marker gene. Accordingly, the clustered regularly interspaced short palindromic repeats associated endonuclease Cas9 (CRISPR/Cas9) system was used to construct a recombinant vector cassette carrying CYP703A3, OrfH79, and DsRed2 that was later transformed into separately developed complete male sterile CYP703A3 mutants to develop a maintainer line. The selfing process of this maintainer line produced both transgenic (fertile) and non-transgenic (sterile) seeds at a 1:1 ratio. The crossing of sterile lines with other elite cultivars (male) produced hybrids with a 13% increase in yield compared to the parental lines (Song et al., 2020). Hence, the application of a CMS gene successfully created a transgene-free plant, outperforming the hybrid plant that is entirely compatible with the “zero tolerance” policy of genetically modified organisms (GMOs) production.

Seed production technology is a great step toward hybrid seed breeding, but not all of the promoters of male specific genes are able to drive the expression of sterility genes, which is a limiting factor for the widespread use of SPT. Strategies should be adopted for the selection of an effective promoter. Focusing on the importance of the use of specific promoters in the SPT-based hybrid production, several methods are being developed to select a strong promoter to drive the expression of pollen killer genes that enable the construction of a SPT maintainer line. For example, Wang M. et al. (2020) used comparative RNA-seq analysis on meiosis-related male sterility genes to select late-stage pollen-specific promoters (LSPs) and identified promoters that could drive the expression of the ZmAA1 gene using DsRed as the marker gene. To harvest the maximum potential of SPT, the identification of promoters thus provides a valuable tool for genetic manipulation of male sterility systems for the production of hybrid rice.

The BARNASE/barstar system uses the BARNASE gene from Bacillus amyloliquefaciens. The gene encodes a toxic ribonuclease RNase and has been used to create dominant male sterility (DMS) (Figure 6A). This system was initially designed for tobacco by fusing the tapetum-specific promoter T29 with the BARNASE gene in order to create male sterility. This strategy was later extended too many crop species including rice, maize, alfalfa, and Brassica (Mariani et al., 1990). The ribonuclease RNase targets tapetum, degrades tapetal RNA, and triggers a premature PCD of the tapetum, resulting in male sterility. The coexpression of BARNASE and barstar reverses the expression of the BARNASE gene and restores fertility in the F₁ generation (Figure 6B). The BARBASE system of transgenic male sterility has achieved lot of success in rice, with a total of 38 anther-specific genes being identified after screening the RiceXPro database (Akasaka et al., 2018). Linking the promoter regions of all of these genes to the barnase gene and expressing it in transgenic rice lines led to male sterility in lines showing expression at early anther developmental stages and exhibited strong flowering habits (Akasaka et al., 2018). The identified genes can be efficiently used to generate male sterile lines and facilitate out-breeding in rice.

The OsC6 is a tapetum-specific gene that encodes a lipid transport protein. The fusion of OsC6’s promoter with the BARNASE gene causes complete male sterility in rice (Kumar et al., 2017). Moreover, the Zm13 is a pollen-specific gene in maize that is expressed in the late stages of pollen development. Fusing the promoter of Zm13 (Zm13pro) with the cytotoxic BARNASE gene and transforming into japonica rice background displayed stable male sterility, which indicates the stable expression of the transgene (Hague et al., 2012). Similarly, the gene BoA9 (from Brassica oleracea L.) is a tapetum-specific promoter that is responsible for male fertility. The linking of BoA9 with the BARNASE gene and subsequent genetic transformation into Nipponbare genetic background resulted in complete male sterility. Furthermore, the open cross-pollination of this sterile male line with fertile plants produced < 1.5% seed setting. In contrast, emasculation and manual pollination resulted in more than 86% seed setting (Abe et al., 2018).

Similarly, in the cysteine protease/cystatin system, the tapetum-specific expression of a cysteine protease impairs the development of the tapetum and causes male sterility (Figure 6A). The Os12bglu38 gene is specifically expressed in the anthers of rice, while the Brassica napus cysteine protease 1 (BnCysP1) is expressed during tapetal PCD and causes male sterility. The fusion of Os12bglu38’s promoter (P1) with BnCysP1 and transformation into a cultivated rice variety resulted in complete male sterility without producing harmful effects on female fertility (Rao et al., 2018). The coexpression of cystatine with cysteine protease system (Shukla et al., 2016) or the crossing of male sterile lines with other transgenic rice lines that were developed by silencing BnCysP1 through RNA interference (RNAi) are able to restore fertility in the F1 generation (Rao et al., 2018) (Figure 6B). Over the past decade, several genes specifically associated with anther development have been identified. The application of these genes in integration with biotechnological approaches can accelerate the production of male sterile lines and hybrid cultivation.

Moreover, the DMS can promote recurrent selection (RS), which is an excellent method to improve the quantitative traits of the population for multiple characteristics. However, the involvement of dozens of parental lines and inefficient crossing techniques has restricted its utilization. The application of RS breeding can be improved by the introduction of GMS in rice (Grenier et al., 2015). The DMS can accelerate the efficiency of RS, but the dominant GMS genes have rarely been found in rice breeding system (Yang et al., 2012; Pang et al., 2017). Hence, DMS through BARNASE/barstar and cysteine protease/cystatin system would facilitate future RS breeding strategy for the development of lines with potential resistance against multiple biotic and abiotic stresses.

The Escherichia coli argE gene deacetylates the non-toxic compound N-acetyl-L-phosphinothricin (N-ac-PPT) and converts it to a toxic herbicide compound that is harmful to the cells and known simply as phosphinothricin (PPT). The argE gene is expressed in tapetum cells of transgenic tobacco and results in male sterility upon treatment with N-ac-PPT because of the accumulation of PPT in the tapetum (Kriete et al., 1996). The promoter of the Oryza sativa indica pollen allergen (OSIPA) contains various cis-regulatory elements and is involved in different anther developmental stages in tobacco, Arabidopsis, and rice (Swapna et al., 2011). The expression of the argE gene under the activity of the OSIPA promoter produced an inducible male sterility system in rice (Figure 6C). Specifically, the transgenic rice plants underwent complete sterility when treated with N-ac-PPT, but this process did not affect female fertility. The restoration of male fertility does not require a fertility restorer line, as argE lines that were not treated with PPT are fertile and can propagate female (sterile) lines (Rao et al., 2017).

The advancement in functional genomics has helped develop high-quality database annotations that can be subjected to next-generation sequencing in order to identify target genes and develop whole-genome-scale mutant libraries. The traditional methods of mutant library development are mainly based on the generation of random mutations via mutagens such as ethyl methanesulfonate (EMS), irradiations, T-DNA insertions, transposons, and chemical treatment. However, obtaining a stable expression of loss-of-function mutations and determining the relationship between genotype and phenotype in these traditional mutants is both time consuming and laborious. Recently, CRISPR/Cas9 has helped establish genome-wide mutant libraries (Lu et al., 2017; Meng et al., 2017). Ma et al. (2019) screened the RiceXPro¹ database using bioinformatic techniques and microarray analysis and identified a total of 1078 genes specifically expressed in the anther. Of these, 555 showed specific expression at the trinucleate stage of anther development, while the remaining 523 genes were expressed before the trinucleate stage. Finally, 73 anther-specific genes that were highly expressed during meiosis and the uninucleate stage of microspore development were knocked out using CRISPR/Cas9 and transformed into japonica rice variety Zhonghua 11 (ZH11) by Agrobacterium-mediated transformation. A total of 15 (out of 73) mutants displayed diversified male sterile phenotypes, including partial sterility, typically abortive sterility, no-pollen sterility, and no-anther sterility (Ma et al., 2019). These observations suggest that the creation of whole-genome rice male sterile mutant libraries offers numerous possibilities for functional genomics research and the development of innovative germplasm resources for breeding hybrid rice in the future.

High-quality hybrid breeding to create resistance against biotic and abiotic factors has always been a preferable approach for the improvement of crop productivity. The availability or generation of a stable male sterile line is the prerequisite to produce high-quality hybrids. Over the years, EGMS lines have been broadly used for two-line systems of hybrid breeding because of (1) availability of germplasm resources and (2) reversibility of fertility via the manipulation of environmental factors (Chen and Liu, 2014; Xue et al., 2018; Zhu et al., 2020; Zhang Y. et al., 2020). On the other hand, abrupt environmental fluctuations have always posed a serious threat for their wider applications. A number of genome editing tools have been designed to manipulate the genetic code. Of these, the CRISPR/Cas9 is the most efficient tool to enable genome editing in plants (Belhaj et al., 2015; Weeks et al., 2016; Zafar K. et al., 2020).

The CRISPR/Cas9 system has been applied to create specific mutations in a thermosensitive gene TMS5 and resulted in the development of 11 new TGMS indica rice lines within one growing year (Zhou et al., 2016). Recently, a very similar experiment used CRIPSR-Cas9 to edit the TMS5 gene and generated two TGMS lines in an indica rice cultivar background (Barman et al., 2019). Moreover, CRISPR/Cas9 multiplex gene editing technology has assisted in pyramiding the TMS5 gene with disease-resistant genes to generate thermosensitive disease resistance male sterile line. For example, the recessive pi21 (Fukuoka et al., 2009) and xa13 (Chu et al., 2006) provide resistance against rice blast and bacterial blight, respectively. The triple tms5, pi21, and xa13 mutant displayed TGMS-like traits and also enhanced resistance against rice blast and bacterial blight (Li et al., 2019). The CSA is a prominent gene for the development of rPGMS-like traits exhibiting male fertility under long-day conditions and male sterility under short-day conditions in japonica rice. The manipulation of CSA by the CRISPR/Cas9 technology developed two reverse PGMS lines, namely, 9522csa and JY5Bcsa, and one rPTGMS145 line, specifically KY131csa-4 (Li Q. et al., 2016). Furthermore, the photo-thermosensitive GMS gene P/TGMS2-1 was cloned from a widely used P/TGMS line Peiai64S. The editing of ptgms2-1 by CRISPR/Cas9 created two new P/TGMS lines in two different indica backgrounds (Lan et al., 2019). Thus, CRISPR/Cas9 can efficiently be applied to manipulate EGMS genes through targeted disruption to produce hybrids with the desirable improved traits.

Multiplex genome editing using CRISPR/Cas9 has the possibility to edit multiple targets simultaneously (Zsögön et al., 2018), which could efficiently assist in pyramiding the several genes of interest along with male sterility genes to develop male sterile lines for broader benefits in rice breeding. Despite the robust benefits of the technology, the use of CRISPR/Cas9 has concerns of off-target mutations in the unintended sequences. In plants, however, such unexpected mega-alterations have not yet been observed, but the probability must be taken into account to avoid deleterious effects of the technology.

More recently, a strategy was proposed to create male sterile lines by disrupting the biosynthesis of jasmonic acid. Specifically, OsOPR7 encodes a 12-oxophytodienoate reductase, which is a precursor for the biosynthesis of JA (Tani et al., 2008). osopr7 mutants created using CRISPR/Cas9 displayed anther indehiscence and complete male sterility. The exogenous application of methyl jasmonate (MeJA) makes it possible to restore fertility and facilitates seed setting through the selfing of mutant plants (Pak et al., 2020). The seeds can then be used for the propagation of male sterile lines and hybrid production by crossing the male sterile lines with desirable fertile lines (Figure 7). Importantly, this male sterility system is insensitive to environmental effects and the application of MeJA is not hazardous to human health. Hence, this system can be widely applied to the production of hybrid rice crops in the future.

The manipulation of OsOPR7 successfully produced phytohormone genic male sterility and then transgene-free hybrid, suggesting its possible application for other genes such as OsFTIP7 because its auxin-deficient mutant and osopr7 mutant have consistent anther indehiscence phenotype, whereas the majority of pollen in both of the mutants remains fertile. Both the mutants exhibit mitotic disorder, which can be reversed through exogenous application of concerned phytohormone. However, the idea may not necessarily work for other phytohormones-related genes. For example, the Gibberellins (GAs)-deficient mutant Osatg7-1 is male sterile characterized by significantly reduced GA contents and defective microsporogenesis. The exogenous application of GAs cannot restore pollen fertility (Kurusu et al., 2017; Hanamata et al., 2020). The different behavior of different phytohormone male sterile mutants suggests that the application of technology will largely depend on the underlying genetic cause of male sterility, hence can be applied to the selected genes, which demands enhanced understandings.

Hybrids are the ultimate source of improved production, but successive selfing and genetic segregation can induce inbreeding depression and the loss of hybrid vigor. During a normal meiotic process, the gametes undergo recombination, crossover, and fertilization to produce recombinant inbred plants (Figure 8A). This process, upon successive selfing, reduces heterosis. Hence, bypassing meiosis and fertilization could directly convert the gametes into adult plants that could be propagated through seeds without segregation in the successive generations through a phenomenon called apomixes. This can be achieved by modulating the postfertilization embryogenesis pathway. A genetic approach called “mitosis instead of meiosis” (MiMe) eliminates recombination by substituting mitosis for meiosis, producing clonal seeds through parthenogenesis (Marimuthu et al., 2011; Mieulet et al., 2016). Clonal propagation of F₁ seeds through apomixes maintains the genome-wide heterozygosity in hybrid plants by avoiding segregation.

The MATRILINEAL (MATL; also known as NOT LIKE DAD and PHOSPHOLIPASE A1 (ZmPLA1) is a pollen phospholipase gene that has been edited to produce haploid maize plants (Gilles et al., 2017). The MATL ortholog in rice is OspPLAIIφ, a gene that has a pollen-specific expression pattern in mature panicles before pollen shedding. Targeted CRISPR/Cas9-induced OspPLAIIφ mutations also resulted in haploid plants in rice that were able to produce normal pollen and seed setting at a low rate. Haploidy analysis revealed a haploid induction rate of up to 1.6% in the E₁ population, which increased to 4.6% in the E₂ population after outcrossing with other female tester lines. On average, the haploid induction rate through homozygous mutation in OspPLAIIφ was ∼6% with an average seed setting rate of up to 20%, which is similar to the results achieved in the maize MATL. This observation led to the renaming of OspPLAIIφ to OsMATL (Yao et al., 2018).

The BABY BOOM1 (BBM1) gene, a member of the APETALA 2 transcription family, can trigger parthenogenesis through its ectopic expression and bypass fertilization in the female gamete. The triple BBM1, BBM2, and BBM3 mutant promotes mitosis instead of meiosis (MiMe) and avoids fertilization, resulting in embryo abortion. This aberrant expression can be fully rescued by male transmitted BBM1 (Figure 8B). The combined expression of MiMe and BBM1 in the egg cells produces diploid maternal clones through parthenogenesis that are able to retain the F₁ heterozygosity (Khanday et al., 2019). Despite the tremendous advantages of producing asexual lines, an apomictic lineage of BABY BOOM genes fosters side effects of polyploidy, which has always been a reproductive barrier through genetic self-incompatibility of the parental lines. The self-fertilization of the diploid clones leads to ploidy doubling in each generation, which should be prevented for true apomixes. The issue has recently been resolved by the introduction of OsMATL gene.

The issue has recently been resolved by the use of rice OsMATL gene. The editing of the pollen-specific rice gene MATRILINEAL (OsMATL) can produce haploid seeds from hybrid rice (Yao et al., 2018). Moreover, the triple mutant of the meiosis-specific genes REC8, PAIR1, and OSD1 turns meiosis into mitosis (MiMe) during female embryogenesis and blocks fertilization to produce an apomictic embryo. Fixing the expression of the haploid inducer OsMATL and the MiMe inducers REC8, PAIR1, and OSD1 through multiplex genome editing results in the elimination of sperm cells and produces diploid seeds by parthenogenesis (Figure 8C). The clonal plants can be propagated through these seeds without undergoing genetic segregation and the doubling of the ploidy level (Wang et al., 2019). This technique is therefore very efficient to engineer meiotic recombination for clonal hybrids with the desired heterozygosity. This mode of reproduction is advantageous compared to the two-line hybrid production scheme and is also less labor intensive. However, it abolishes genetic variation, as the resultant clonal embryos harbor complete maternal genetic complement.

Normal meiosis believes in crossover recombination, a crucial process in plant breeding to create novel allele combinations on chromosomes for breeding superior F₁ hybrids. Silencing the crossover formation/homologous chromosomes pairing is essential for breeders to effectively engineer the chromosomes behavior for hybrid vigor (Wijnker and Jong, 2009). The F₁ heterozygotes cannot be stably propagated through seeds, and establishing breeding lines for elite heterozygote has always been a major obstacle for plants breeders. Reverse breeding can efficiently develop the homozygous parental double haploid lines, which can stably be propagated indefinitely. These homozygous parental lines upon mating can then reconstitute the elite heterozygote. The technology is based on the silencing of meiosis-related target genes in heterozygote, which then produces reduced number of viable haploid. The perfectly homozygous double haploid plants can be regenerated form haploid spores (Dirks et al., 2009; Ren et al., 2017). The Arabidopsis DMC1 has successfully been silenced to develop achiasmatic haploid and double haploid plants to reconstitute the heterozygote (Wijnker et al., 2012, 2014). Over the past decade many homologous recombination-related genes have been identified, but their potential for hybrid rice breeding has never been harvested. For example, the PAIR2, PAIR3, BVF1, and SPO11 are essential for chromosome recombination, and the mutations cause complete male sterility (Nonomura et al., 2004; Yuan et al., 2009; Yu et al., 2010; Zhou et al., 2017). Interestingly, their mutants produce intact univalent, hence perfectly useful. These and the other related genes can be the soft target of knocking down the gene expression through RNAi to promote reverse breeding in rice in the future. However, chromosome recombination can also be engineered to increase crossover incidences and position over chromosomes to develop F₁ hybrids with maximum heterozygosity.