GhGPAT12/25 Are Essential for the Formation of Anther Cuticle and Pollen Exine in Cotton (Gossypium hirsutum L.)

Glycerol-3-phosphate acyltransferases (GPATs), critical for multiple biological processes like male fertility, have been extensively characterized. However, their precise functions and underlying regulatory mechanism in cotton anther development are unclear. This research demonstrated the importance of GhGPAT12/25 (a paralogs pair on A12/D12 sub-chromosome of cotton) to regulate the degradation of tapetum, anther cuticle formation, and pollen exine development. GhGPAT12 and GhGPAT25 exhibited specifically detected transcripts in tapetum and pollen exine during the early anther developmental stages. GhGPAT12/25 are sn-2 glycerol-3-phosphate acyltransferases and can transfer the acyl group of palmitoyl-CoA to glycerol-3-phosphate (G3P). CRISPR/Cas9-mediated knockout identified the functional redundancy of GhGPAT12 and GhGPAT25. Knockout of both genes caused completely male sterility associated with abnormal anther cuticle, swollen tapetum, and inviable microspores with defective exine and irregular unrestricted shape. RNA-seq analysis showed that the loss of function of GhGPAT12/25 affects the processes of wax metabolic, glycerol monomer biosynthesis, and transport. Consistently, cuticular waxes were dramatically reduced in mutant anthers. Yeast one-hybrid system (Y1H), virus-induced gene silencing (VIGS), and dual-luciferase (LUC) assays illustrated that GhMYB80s are likely to directly activate the expression of GhGPAT12/25. This study provides important insights for revealing the regulatory mechanism underlying anther development in cotton.


INTRODUCTION
In flowering plants, male reproductive development is a complex biological process involving the differentiation of stamen primordium to produce anther, the transition from sporophytic to gametophytic generation, and the formation of pollen grains (Scott et al., 2004). The formed anther consists of four somatic layers (from the surface to the interior): the epidermis, endothecium, middle layer, and the tapetum (Goldberg et al., 1995). Microspore mother cells localized at the center of the anther lobes will differentiate and undergo meiosis to produce microspores, which then give rise to pollen grains. Pollen development is intimately linked to the morphogenesis of anther and requires the synergistic interaction between sporophytic and gametophytic genes (Goldberg et al., 1995;Scott et al., 2004).
To ensure the biological processes of microspore development and pollen grain maturation, plants establish two barriers in order to resist various environmental stresses. One of the barriers is the anther cuticle continuously coating the outermost surface of anther epidermis (Piffanelli et al., 1998). Similarly to the epidermal cuticle of vegetative organs, anther cuticle is a complex biopolymer composed of two types of lipophilic materials: cutin and waxes (Kolattukudy, 2001). The hydrophobic cutin matrix is synthesized by hydroxyl and epoxy fatty acids, and it determines the framework of the anther cuticle (Kolattukudy, 2001;Heredia, 2003). Waxes contain different substances generated from (very)long-chain fatty acids through alcohol/alkane-forming pathways, and they are embedded or/and overloaded on the surface of cutic matrix (Kunst and Samuels, 2003). The synthetic pathway of the cuticle involves the synthesis, transport, and polymerization of cutin monomers. Genetic and biochemical evidences have demonstrate that C16 and C18 fatty acids are the main substrates for cutin formation, and they are firstly esterified to produce (fatty) acyl-CoA (Kunst and Samuels, 2003;Zhang et al., 2018). This formation process is catalyzed by long-chain acyl-CoA synthetases (LACS), like LACS1 and LACS2 in Arabidopsis (Lu et al., 2009). Then, the products are oxidized by CYP86 and CYP77 subfamilies members, such as the CYP703A2/CYP703A3 and CYP704B1/CYP704B2 in Arabidopsis/rice (Morant et al., 2007;Dobritsa et al., 2009;Li et al., 2010;Yang et al., 2014). Furthermore, the acyl group of acyl-CoA is transferred to glycerol-3-phosphate (G3P) under the catalysis of glycerol-3-phosphate acyltransferase (GPAT), thus yielding monoacylglycerol cutin monomers (Li-Beisson et al., 2013). OsGPAT3 and ZmMS33 were reported to be genes involved in this aspect for anther cutin formation (Men et al., 2017;Zhang et al., 2018). ATP-binding cassette (ABC) transporters (i.e., ABCG11 and ABCG12) are then recruited for the channeling of glycerol monomers through the plasma membrane (Pighin et al., 2004;Panikashvili et al., 2007). Subsequently, the hydrophobic polymer compounds will pass through the apoplastic compartment or cell wall; lipid transfer proteins (LTPs) collaborate in the completion of this process (Edstam and Edqvist, 2014). Pollen wall, especially the exine, is the second barrier practically for pollen grain maturation and it is also indispensable for pollen-stigma communication (Piffanelli et al., 1998). The main component of the exine is sporopollenin (an extremely chemically inert biopolymer), which was reported to be likely synthesized by polyvinyl alcohol units and 7-O-p-coumaroylated C16 aliphatic units in pine . The synthesis and secretion of the major lipidic precursors of cutin and sporopollenin largely rely on the developing tapetum. Mutations in genes responsible for the formation or transport of these lipid components usually cause abnormal programmed cell death (PCD) in the tapetum. Examples of such genes are AMS/TDR, MYB80, MS1/PTC1, MS2/DPW, CYP703A2/CYP703A3, and CYP704B1/CYP704B2 (Wilson et al., 2001;Sorensen et al., 2003;Li et al., 2006Li et al., , 2010Li et al., , 2011Morant et al., 2007;Dobritsa et al., 2009;Chen et al., 2011;Phan et al., 2011;Shi et al., 2011;Yang et al., 2014;Pan et al., 2020).
In plants, the development of sporophytic anthers and internal pollen grains is considered to be a complex biological process that is controlled by a precisely transcriptional regulatory network involving a series of genes (Feng et al., 2012). Based on mutant and interaction analysis, a number of principal genes controlling this process have been identified in Arabidopsis. Such genes include DYT1, AMS, TDF1, MYB80, MS1, MS2, LAP5, LAP6, CYP703A2, and CYP704B1 (Wilson et al., 2001;Sorensen et al., 2003;Zhang et al., 2006;Morant et al., 2007;Zhu et al., 2008a;Dobritsa et al., 2009;Kim et al., 2010;Chen et al., 2011;Phan et al., 2011). With homologous genes characterized in rice or/and other plants, further studies implied that these genes are evolutionarily conserved. MYB80 is a key transcription factor located downstream of AMS to control the tapetum development and pollen exine formation. MYB80 was reported to directly or indirectly regulate the expression of many functional genes being responsible for the synthesis or transport of lipidic components required for anther development (Phan et al., 2011;Xiong et al., 2016;Lu et al., 2020). Genes, such as CYP703A2, MS2, LAP5, and LAP6, show a significant down-regulation in the MYB80 mutant. Further research has found that MYB80 protein could induce the expression of these genes, thereby supporting the proposal that MYB80 acts as a pivotal connector for upstream TFs and downstream enzyme genes during anther development (Phan et al., 2011). Notably, GhMYB80 can rescue the male sterile phenotype of AtMYB80 in Arabidopsis (Xu et al., 2014), indicating its potential role for anther development in cotton.
The sn-Glycerol-3-phosphate O-acyltransferase (GPAT, EC2.3.1.15) is a key enzyme responsible for the synthesis of glycerophospholipids and triacylglycerol (TAG) (Zheng et al., 2003). Previous studies revealed that diverse GPATs may catalyze the transfer of an acyl group from acyl-CoA/acyl-ACP to the different positions (sn-1 or sn-2) of glycerol-3-phosphate (G3P) (Zheng and Zou, 2001). In Arabidopsis, 10 GPATs (ATS1 and AtGPAT1-9) have been identified. ATS1 and AtGPAT9 possess sn-1 acyltransferase activity and they have been reported to contribute to the synthesis of storage oil (Nishida et al., 1993;Gidda et al., 2009). AtGPAT1-8 are sn-2 acyltransferases likely involved in the biosynthesis of extracellular lipids barrier polyesters, such as cutin and suberin . AtGPAT4 and AtGPAT8 are required for the accumulation of cutin monomers in leaves and stems (Li et al., 2007). AtGPAT5/7 predominantly contribute to the synthesis of suberin, while, AtGPAT7 has also a unique role in wounding response . AtGPAT1 and AtGPAT6 have been reported to be essential for pollen development and largely redundant in tapetal development. Both gpat1 and gpat6 mutants exhibit abnormal tapetal cells and reduced mature pollen grains, while the gpat1 gpat6 double mutant shows short filaments and a complete male sterility (Zheng et al., 2003;Li et al., 2012). Moreover, OsGPAT3 was found to play a crucial role in anther cuticle formation and pollen exine development in rice (Men et al., 2017). Its homologous gene in maize, ZmMS33, also benefits male fertility (Zhang et al., 2018;Zhu et al., 2019). Despite the importance of understanding the roles of GPATs, few relative researches have been performed in cotton. To date, the GPAT family members have been identified in four species: G. raimondii, G. arboreum, G. hirsutum, and G. barbadense. Several probably storage oil-related genes have also been found (Cui et al., 2019). However, the contribution of GPAT genes in cotton male sterility has not been investigated.
In this study, the pivotal role of GhGPAT12/25 in synthesizing the glycerol monomers for anther cuticle formation and pollen exine development in cotton has been investigated. GhGPAT12/25 are sn-2 glycerol-3-phosphate acyltransferases and can transfer the acyl group of palmitoyl-CoA to G3P. As paralogs pair, GhGPAT12 and GhGPAT25 have similar temporal expression pattern and are restricted in tapetum and pollen exine during the early anther developmental stages. Moreover, CRISPR/Cas9-mediated knockout of GhGPAT12/25 caused a male sterile phenotype. Then, it was shown that GhMYB80s are likely to directly activate the expression of GhGPAT12/25. This work determined that GhGPAT12/25 participate in the conserved pathway to synthesize lipidic monomers required for anther cuticle formation and pollen exine development in cotton.

Plant Materials
The upland cotton TM-1 was used in this study.

DNA, RNA, and cDNA
Total DNA of TM-1 anthers was extracted using the CTAB method (Porebski et al., 1997). Total RNA of various tissues and anthers of different developmental stages were isolated using the RNAprep Pure Plant Kit (Tiangen, Beijing, China). The firststrand cDNA was prepared using the PrimeScript RT reagent kit with a gDNA Eraser (Takara, Dalian, China), following the manufacturer's instructions.

qRT-PCR
qRT-PCR analyses were carried out using SYBR Premix Ex Taq (Tli RNaseH Plus) (Takara, Dalian, China). Gene-specific primers were directly obtained from qPrimerDB 2 . The cotton ACTIN 7 was used to as an internal control gene. The Ct algorithm was used for calculating relative gene expression. All primers used in this study are presented in Supplementary Table 3.

RNA in situ Hybridization
TM-1 anthers at various developmental stages were sampled and fixed in formalin/acetic acid/alcohol fixative solution for 12 h at 4 • C. After being dehydrated through the ethanol series, anther samples were embedded in paraffin and sectioned to a thickness of ∼7 µm using a rotary microtome (RM2016, Leica). The 29-bp specific fragment corresponding to the GhGPAT12/25 cDNA was used to design the antisense and sense probes. Digoxigenin-labeled RNA probes were labeled using DIG-UTP (Roche). Details of RNA hybridization and signal detection of the hybridized probes are described by Kouchi and Hata (1993).

Bioinformatic Analysis of GhGPAT12/25
The DNA and CDS sequences of GhGPAT12/25 were cloned from TTP stage anthers of TM-1. The basic characters of GhGPAT12/25 proteins were analyzed using ProtParam 3 . Their conserved domains were predicted using Pfam 4 . The 26 corresponding amino acid homologous sequences from 10 different species were obtained from NCBI 5 and TAIR 6 . Multiple sequence alignment of GhGPAT12/25 and homologous sequences of other species were performed using DNAMAN. A neighbor-joining phylogenetic tree was constructed using MEGA 6.0 7 and 1000 repetitions of bootstrap analysis were performed.

CRISPR/Cas9-Mediated Knockout
The full-length DNA sequences of GhGPAT12/25 were analyzed using CRISPR-P software 8 , and two sgRNA targets were selected to assemble into the pRGEB32-GhU6.7-NPT II expression vector. Then, the recombinant vector was transferred to Agrobacterium tumefaciens (GV3101). Upland cotton cv. HM-1 was used as the transformation receptor. The transgenic plants were obtained following the method described by . The HI-TOM analysis described by Liu et al. (2019b) was used to check the sequences of transgenic plants.
WT and mutant were sampled for paraffin section, SEM, TEM, and TUNEL. The detailed methods were as described by Uzair et al. (2020), Zhang et al. (2020).

Wax Determination
WT and mutant anthers were collected during the mature stage. The surface areas of cotton anthers in WT and mutant were determined using microscopy. For waxes quantification, 100 mg of freeze-dried anther was submersed in 10 ml of chloroform (55 • C) for 1 min. Extraction was repeated once, and the chloroform extracts were spiked with 100 µg of C24 Alkan. The solvents were evaporated under nitrogen gas. The remaining compounds were incubated with 200 µl of BSTFA (Sigma-Aldrich) and 200 µl of pyridine for 40 min at 70 • C. These derivatized samples were then analyzed using GC-MS and GC-FID.

Yeast One-Hybrid Assay
The promoter sequences of GhGPAT12/25 were analyzed using New PLACE 9 to identify specific cis-element motifs bound by upstream genes. Two specific promoter segments were synthesized for three copies and constructed into the pHIS2 vector. The CDS sequences of GhDYT1-A (GH_A10G0179), GhAMS-A (GH_A12G0356), GhMYB80-A (GH_A04G0015), GhTDF1-A (GH_A13G1049), and GhbHLH91-A (GH_A06G1836), potential upstream transcription factors that showed similar expression pattern with GhGPAT12/25, were cloned into the pGADT7 vector. Each pHIS2 bait vector was co-transformed with recombinant pGADT7 prey vector to Y187 yeast strain. The co-transformed yeasts were incubated on SD/-Leu/-Trp selective medium and further identified on SD/-Leu/-Trp/-His selective medium with 200 mM 3-amino-1,2,4-triazole (3-AT). The combinations of pGADT7 empty vector and recombinant pHIS2 vectors were employed as negative controls.

Dual-Luciferase Assay
A transient dual-luciferase assay in tobacco leaves was carried out to confirm the activation effect of GhMYB80s on GhGPAT12/25. The promoter sequence of GhGPAT12 was amplified and cloned into the pGreenII0800-LUC vector as the reporter plasmid. The CDS of GhMYB80-A was cloned into the pGreenII 62-SK vector as effecter plasmid. The recombinant vectors were then transformed into the GV3101 (pSoup-p19) strain. Subsequently, equal A. tumefaciens cells that carried the reporter and effector vectors were co-infiltrated into tobacco leaves. Two days later, the infected tobacco leaves were sampled for LUC and REN luciferase activity analysis. To achieve this, the Dual-Luciferase Reporter Assay System (Promega, Madison, WI, United States) and GloMax 20/20 Luminometer (Promega) were used. Moreover, the pGreenII0800-LUC-pGhGPAT12 and pGreenII 62-SK (empty) vectors were used as a negative control, and each combination was replicated six times. 9 http://bioinformatics.psb.ugent.be/webtools/plantcare/html/

VIGS Assay
To further illustrate the potential function of GhMYB80 in cotton anther development, a VIGS assay was performed using upland cotton TM-1 as receptor. Briefly, a 300-bp fragment of the GhMYB80-A CDS located outside of the conserved region was cloned and constructed into the pCLCrVA vector. The specific primers are listed in Supplementary Table 3. The empty pCLCrVA and pCLCrVA-PDS were used as negative and positive controls, respectively. These plasmids were transferred into A. tumefaciens (strain GV3101). The A. tumefaciens cultures containing the helper vector (pCLCrVB) and pCLCrVA or its derivative (pCLCrVA-PDS and pCLCrVA-GhMYB80s) vectors were mixed at ratio of 1:1 and then co-injected into the fully expanded cotyledons of c. 10-day-old TM-1 seedlings. The plants were cultivated at a 23 • C phytotron with 16 h:8 h light:dark photoperiod. At the flowering stage, the phenotype of pCLCrVA-GhMYB80s individuals were observed for the identification of male sterile plants, and the TTP stage anthers of different plants were collected for qRT-PCR analysis.

RNA-Seq
A total of eight libraries, comprising two cotton groups (ghgpat12/25 and WT) at two stages (TTP and early UNP stage) and two biological replicates, were sequenced using Illumina NovaSeq 6000 to generate paired-end reads. Lowquality reads were first removed, and the resulting clean reads were aligned with the G. hirsutum (TM-1) genome (Hu et al., 2019). StringTie was used to accurately quantify the fragments per kilobase of transcript per million mapped reads (FPKM) for calculating gene expression levels. Differential expression genes were obtained using the DESeq R package. Genes with the | log2 fold change| ≥ 1 and FDR ≤ 0.01 were considered to be differentially expressed. Gene Ontology (GO) enrichment analysis of DEGs was performed using OmicShare tools 10 and the ClueGO plugin in Cytoscape 3.3.0. STRING 11 was used to identify the known and predicted interactions in DEGs. Then, the ClueGO plugin was used for the enrichment analysis of PPI network-containing genes. RNA-seq raw data are accessible in the NCBI Sequence Read Archive (SRA) database under Accession Number PRJNA698752.

GhGPAT12/25 Have Early Anther Stage-Specific Expression in Cotton
Previous RNA-seq data of anther samples at the tetrad pollen (TTP), uninucleate pollen (UNP), and binucleate pollen (BNP) stages are important references to identify crucial genes for anther development (Zhang et al., 2020). Public expression data provide an additional rich resource of information on multiple tissues expression of cotton genes (Zhang et al., 2015). Based on the combination analysis of these data, GhGPAT12/25 (glycerol-3-phosphate acyltransferase genes located on A12/D12 chromosomes) were found expressed in anthers but not in any other tissues, and they also showed strict expression during TTP stage ( Figure 1A).
To confirm this, qRT-PCR analysis was performed using anthers at the developmental stages of microspore mother cell (MMC), meiosis cell (MC), tetrad pollen (TTP), early uninucleate pollen (eUNP), late uninucleate pollen (lUNP), binucleate pollen (BNP), mature pollen (MP), and vegetative tissue samples of cotton plants. The results suggested that GhGPAT12/25 were anther-specific expressed and showed negligible expression in all the other tissues tested. In anther tissues, the expression of GhGPAT12/25 was detectable as early as the MMC stage, peaked at the MC stage, declined gradually at the TTP stage, and showed a lower level after the eUNP stage. In addition, GhGPAT12 showed a similar expression pattern to GhGPAT25, but its expression levels were higher at MC and TTP stages than those of GhGPAT25 ( Figure 1B).
RNA in situ hybridization with anther sections was performed to further precisely determine the spatial and temporal patterns of GhGPAT12/25 transcripts. GhGPAT12/25 signal was initially detected in tapetal layer at the MMC stage ( Figure 1C), and then a strong expression signal was detected predominantly in the tapetum and microspores from the MC to the TTP stage ( Figures 1D,E). From the TTP stage onward, the signal of GhGPAT12/25 transcripts decreased in the tapetum but showed relatively high abundance in microspores exine, especially on the spines ( Figure 1F). Subsequently, the signal was undetectable in the BNP stage ( Figure 1G). Meanwhile, no expression signals were detected by the sense probe in anthers sections during the UNP stage ( Figure 1H).
This expression pattern implied that GhGPAT12/25 may play specific roles in microspore development and pollen wall formation.

GhGPAT12/25 Are Active sn-2 Glycerol-3-Phosphate Acyltransferases Conserved in Land Plants
GhGPAT12/25 were cloned from the TTP stage anther DNA/cDNA of upland cotton TM-1. The genomic DNA sequence was 1976/1977 bp in length, containing two exons and one intron (Figure 2A). The coding region was 1626 bp, and the corresponding 541-aa protein had a molecular mass of 60.99/60.98 kDa and an isoelectric point of 9.12/9.29.
GhGPAT12/25 were glycerol-3-phosphate acyltransferases and were predicted to contain a conserved acyltransferase domain (Pfam: PF01553) at the C-terminal region. To understand the evolutionary roles of GhGPAT12/25 and gain information on their potential function, the full-length proteins were used as queries to search their homologs in two public databases: National Center for Biotechnology Information (NCBI see text footnote 5) and The Arabidopsis Information Resource (TAIR see text footnote 6). This search yielded a total of 26 close homolog protein sequences from 10 different species. Subsequently, a phylogenetic tree was constructed, and the 28 sequences were grouped into three main clades ( Figure 2B). AtGPAT1-8 were reported to belong to sn-2 GPATs, which contribute to the formation of polyester, such as cutin and suberin . The distribution of these Arabidopsis members in three different groups of the phylogenetic tree indicated that homologous proteins in these different species may have similar functions with AtGPAT1-8.
GhGPAT12/25, together with their orthologs from Hibiscus syriacus, Durio zibethinus, Theobroma cacao, Herrania umbratical, Corchorus capsularis, Populus trichocarpa, Nicotiana tabacum, Zea mays, and Arabidopsis, were grouped in clade I. OsGPAT1 and all GPAT2/3 proteins were clustered in clade II, while clade III included the remaining sequences ( Figure 2B). Sequence alignment showed that GhGPAT12 and GhGPAT25 share an acyltransferase (AT) domain with similar conserved boxes (AT-1 to AT-IV) containing the corresponding catalytic residues and binding residues with 26 other proteins ( Figure 2C). With the exception of the AT region, haloacid dehalogenase (HAD)-like domains containing phosphohydrolases were recognized at the N-terminal region of GPAT. According to the protein sequence analysis, except for AtGPAT5 and AtGPAT7, other proteins of clade III had the typical DXD signature and GDXXXD motif required for phosphatase activity ( Figure 2C). However, other 20 proteins lacked these intact residues (Figure 2C), indicating their absent phosphatase activity. Furthermore, members of clade III, such as AtGPAT4-8, were involved in the formation of polyester in vegetative organs and the improvement of tolerance to environment stresses . Proteins in clade I or II, like AtGPAT1, OsGPAT3, and ZmMS33, have been shown to be essential for pollen development (Zheng et al., 2003;Men et al., 2017;Zhang et al., 2018;Zhu et al., 2019; Figure 2C). These analyses suggested the functional differentiation of genes from different clades in vegetative and reproductive development. Altogether, these observations indicated a conserved structure of GhGPAT12/25 in sn-2 glycerol-3-phosphate acyltransferase and potential roles in plant male reproductive development.
To determine whether GhGPAT12/25 have the activity to acylate glycerol-3-phosphate, GhGPAT12 and GhGPAT25 were cloned into pYES2 vector; both proteins were heterologously expressed in the yeast strain gat1 , which harbors a mutant ERbound GPAT gene with very low GPAT activity (Zheng and Zou, 2001). Under the control of the galactose-inducible promoter of GAL1, the GhGPAT12 and GhGPAT25 proteins were expressed, and total yeast homogenates were prepared for the detection of glycerol-3-phosphate acyltransferase activities with palmitoyl-CoA as the fatty acyl donor. Thus, overexpression of GhGPAT12 and GhGPAT25 resulted in an increase of about 617 and 470 pmol·min −1 ·mg −1 in glycerol-3-phosphate acyltransferase activities, respectively ( Figure 2D). The significant increases of specific activities in yeast proved directly that GhGPAT12 and GhGPAT25 indeed encoded active glycerol-3-phosphate acyltransferases.
CRISPR/Cas9-Mediated Knockout of GhGPAT12/25 Leads to Male Sterility CRISPR-Cas9 was performed in upland cotton "HM-1" to knock out GhGPAT12/25 genes in order to validate their role in reproductive development. Because of the high similarity of sequence and temporal expression pattern of GhGPAT12/25 (Figure 1 and Supplementary Figure 1), both genes were targeted for knockout. Two sgRNAs were designed targeting the 1st exons (323 bp away from each other) and cloned into a single Cas9-sgRNA cassette ( Figure 3A).
Eleven independent T 0 transgenic lines were acquired. Hi-TOM (Liu et al., 2019b) was performed to evaluate the editing efficiency and to precisely analyze the nucleotide insertion and deletion of GhGPAT12/25. The results showed that all 11 transgenic lines were positive plants with various types of genome editing events precisely occurring at the examined gene, GhGPAT12 or/and GhGPAT25 ( Figure 3B and Supplementary  Figure 2). Among such transgenic lines, T 0 -2, T 0 -12, T 0 -15, T 0 -23, T 0 -25, and T 0 -26 possessed both edited GhGPAT12 and GhGPAT25 sequences at the target site, but no normal WT sequences ( Figure 3B and Supplementary Figure 2). For instance, T 0 -23 exhibited three types of indels at two target sites in GhGPAT12 and two types of indels at sgRNA2 site in GhGPAT25 (Supplementary Figure 2). However, T 0 -1, T 0 -6, T 0 -7, T 0 -11, and T 0 -28 were incompletely edited lines that contained WT sequences in either GhGPAT12 or GhGPAT25 ( Figure 3B and Supplementary Figure 2). Among these, the T 0 -1 and T 0 -28 plants presented WT sequences of both GhGPAT12 and GhGPAT25, while the T 0 -6 and T 0 -11 plants possessed several types of indels on At and Dt sub-genes and held WT sequences of GhGPAT12 (Figure 3B and Supplementary Figure 2). T 0 -7 showed a completely edited GhGPAT12 gene with three types of indels, but incomplete GhGPAT25 with one type of indel and the WT sequences ( Figure 3B). To evaluate the function of GhGPAT12/25, the T 0 and T 2 (F 1 self-pollinated progeny of T 0 × HM-1) transgenic lines were planted in agronomic field for morphological analysis. The results showed that the incompletely edited lines T 0 -1, T 0 -6, T 0 -7, T 0 -11, and T 0 -28 had a normal phenotype with fertile anthers. Whereas the completely edited lines, such as T 0 -15, exhibited normal vegetative growth similar to HM-1 (Figures 3C,D), the reproductive organs were defective ( Figure 3F). The flower of T 0 -15 displayed shorter filaments with shriveled anthers (Figure 3F). The anthers did not dehisce and lacked mature pollen grains at the late developmental stages when stained with 1% I 2 -KI solution (Figures 3F,H). T 2 plants of T 0 -15 showed fertility segregation; according to the results of the Hi-TOM analysis, the F 2 progenies with homozygous or biallelic mutations of GhGPAT12/25 were completely male sterile (Supplementary  Figure 3), while the genotypes of fertile plants were homozygous wild type or heterozygous (Supplementary Figure 3). These results clearly indicated that GhGPAT12 and GhGPAT25 play key roles in the normal development of male organs in cotton and appear to be functionally redundant.

GhGPAT12/25 Contribute to the Assembly of Anther Cuticle and Pollen Exine
To investigate the cytological effects of GhGPAT12/25, histological transverse section analysis of WT and ghgpat12/25 (a T 2 male sterile plant) anthers was employed. No significant differences between ghgpat12/25 and WT were detected at the MMC stage. Their anthers formed four typical anther wall layers and their microsporocytes were located at the center of each anther locule (Figures 4A,B).
Subsequently, ghgpat12/25 anthers displayed clear morphological abnormalities. At the MC stage, WT tapetal cells started to shrink and showed deep staining with toluidine blue (Figure 4C), while the ghgpat12/25 tapetal cells did not appear to condense and were weakly stained (Figure 4D). At the TTP stage, anther wall layers of WT anthers became thinner, tetrads of haploid microspores were formed in anther locule, and the tapetal layer was more condensed (Figure 4E). In contrast, the wall layers of ghgpat12/25 anthers seemed to be irregular and thicker, the microspores in ghgpat12/25 tetrads exhibited an irregular shape, and the tapetal layer was swollen and lightly stained (Figure 4F). At the UNP stage, WT microspores covered with spines were released from the tetrads. The tapetal layer underwent PCD and apparently decayed into a thinner layer ( Figure 4G). However, the microspores of ghgpat12/25 showed an anomalous appearance without spines ( Figure 4H). Meanwhile, the ghgpat12/25 tapetum was still pachytic (Figure 4H). At the BNP stage, the WT anther locules continued to expand with thinner wall layers, microspores became spherical with large vacuoles, and the tapetum was almost completely degraded into cellular debris (Figure 4I). By contrast, ghgpat12/25  displayed narrow anther locules with thicker wall layers and tapetum, as well as microspores without large vacuoles and abnormal appearance despite the presence of the expansion (Figure 4J). From the vacuolated pollen stage to the mature pollen stage, the tapetum of the WT gradually degenerated and vacuolated microspores turned to mature pollen grains with fully accumulated nutrients (Figures 4I,K). However, the tapetum of ghgpat12/25 seemed to undergo acute and abnormal degradation, leaving only aborted pollen grains in the atrophic anther locules (Figures 4J,L).
Scanning electron microscopy (SEM) was used to further investigate the morphological differences of anthers and pollen grains in WT and ghgpat12/25 at the late development stages. The WT anther exhibited a plump and normal cuticle-covered surface (Figures 5A,B). The pollen grains were covered by exine pattern full of spines and apertures (Figures 5C,D). In ghgpat12/25, a shriveled and relatively smooth anther epidermis (Figures 5E,F) and shrunken pollen grains were produced as a result of the disorganized cuticle structures and the anomalously assembled exine (Figures 5G,H).
To more precisely characterize the defects in ghgpat12/25, transmission electron microscopy (TEM) analysis was performed on the anthers from the TTP stage to the BNP stage. At the TTP stage, the tapetum showed a thin cell wall layer, the tapetal cells became largely vacuolated, and the middle layer began to degenerate ( Figure 6A). Additionally, rounded microspores were observed within the tetrad (Figure 6B). Unlike those tapetum of WT plants, ghgpat12/25 tapetum was disordered and its middle layer was much thicker (Figure 6C). At the same stage, although the tetrads of ghgpat12/25 were also covered by callose, the microspores seemed to develop abnormally with irregular surface (Figure 6D). At the early UNP stage and within the tapetal cells of WT, the vacuoles were reabsorbed and the tapetum started to degenerate ( Figure 6E). Meanwhile, the elementary structure of the exine, composed of the well-organized tectum, bacula, and nexine, was observed at the surface of the WT microspores being released from the tetrads (Figures 6F,G). Conversely, in the ghgpat12/25 anthers, the middle layer persisted and became largely vacuolated ( Figure 6H); the tapetal cells seemed slightly stained and became abnormally expanded, which caused one side to be in direct contact with the middle layer ( Figure 6H). The ghgpat12/25 microspores displayed an irregular shape (Figure 6I), and the pollen exine seemed to contain relatively normal nexine, but the tectum and bacula were poorly developed and lacked the assembly of spines (Figure 6J). At the late UNP stage, the WT microspore underwent large vacuolation period, the tapetum continued to degenerate, and the sporopollenin was deposited onto the microspore surface thickening the exine (Figures 6K-M). However, a stunted middle layer and swollen tapetum cells with intact nuclei were observed in the ghgpat12/25 anther ( Figure 6N). The ghgpat12/25 pollen nexine did not show significant thickening, and the assembly of sexine was still abnormal (Figures 6O,P). At the BNP stage, the WT anther showed an almost completely degenerated tapetal cell layer ( Figure 6Q) and spherical pollen grains with accumulated starch and lipidic materials ( Figure 6R). Furthermore, the pectocellulosic intine started to develop beneath the exine, and the tryphine was deposited onto the pollen wall ( Figure 6S). However, ghgpat12/25 anther showed no obvious degeneration of the middle layer but a vanished tapetal layer (Figure 6T), leaving aborted microspores with thin intine and absent tryphine in anther locules (Figures 6U,V).
In synthesis, it can be concluded that GhGPAT12/25 are required for the development of an anther cell layer, tapetum PCD, pollen wall assembly, and anther cuticle formation.

DNA Fragmentation in ghgpat12/25 Tapetal Cells Is Abnormal
The timely degradation of tapetal cells caused by the process of PCD is considered to be vital for anther cuticle formation and pollen development. Paraffin section observation and TEM suggested that ghgpat12/25 tapetal cells abnormally expanded with the development of pollen. To examine whether ghgpat12/25 anthers are defective in the tapetum PCD process, a TUNEL (terminal deoxynucleotidyl transferase-mediated dUTP nickend labeling) assay was performed. DNA fragmentation was not detected at the MC stage in both WT and ghgpat12/25 tapetal. Tapetal DNA fragmentation signals were firstly visualized in the WT anthers at the TTP stage (Figures 7A,F). As the tapetum layer continued to degenerate, positive TUNEL signals continued to increase until reaching a maximum at the UNP stage ( Figure 7B), and then the signals declined and almost disappeared after the BNP stage (Figures 7C,D). However, no clear DNA fragmentation was detected in ghgpat12/25 anthers, with the exception of the few signals scattered across the tapetum at the UNP stage (Figures 7G-J). These TUNEL assay results further indicated that the precise PCD of ghgpat12/25 tapetum was disrupted.

Loss of Function of GhGPAT12/25 Affects the Expression of Genes Responsible for Lipid Synthesis and Transport
To further gain insight into the functional mechanism of GhGPAT12/25 in the control of anther development, RNA-seq of anthers at the TTP and early UNP stage was established to compare the global gene expression of ghgpat12/25 and WT plant. A total of 3655 original differentially expressed genes (DEGs) were filtered with the | log 2 fold change| ≥ 1 and FDR ≤ 0.01. Of these genes, 1111 were downregulated (uninvolved in up-regulation in any stages) and 2315 were up-regulated (uninvolved in down-regulation in any stages). Enrichment analysis of DEGs showed that the down-regulated genes were involved in "polyol transport, " "transmembrane transport, " "lipid biosynthetic process, " "glycerol transport, " and "wax metabolic process, " among others (Figure 8A), whereas the up-regulated genes were mainly enriched ( Figure 8A) to be responsive to various stimuli, such as hormone, chemical, chitin, etc.
Notably, the down-regulated genes included CER1 (GH_A06G1828 and GH_D06G1860), CER3 (GH_A13G2356), and CER6 (GH_A01G2012 and GH_D01G2106) ( Supplementary  Table 1), which have been reported to be male sterile genes contributing to the biosynthesis of cuticle membrane and wax in Arabidopsis (Aarts et al., 1995;Fiebig et al., 2000;Ariizumi et al., 2003). Moreover, other wax and cuticle pathway involving genes, such as LACS1 (Lu et al., 2009) (GH_A07G1407), CER2 (Haslam et al., 2012, and SHN1 (Broun et al., 2004) (GH_A10G0747), also exhibited lower expression in ghgpat12/25 (Supplementary Table 1). For the assembly of anther cuticle, lipidic monomers synthesized by glycerol-3phosphate acyltransferase are required to be transferred to the epidermis layer by LTPs and ABC transporter proteins (ABCGs) (Li-Beisson et al., 2013). Here, many LTPs and ABCGs genes were found to be down-regulated in the mutant (Supplementary Table 1); among such genes are LTPG1 (GH_A04G1689 and GH_D04G2037), LTPG6 (GH_D05G0009 and GH_D08G0861), LTP4 (GH_A07G0829), LTPG26 (GH_D02G0373), ABCG14 (GH_D08G1774), and ABCG25 (GH_D05G2788). Subsequently, the decreased expression of these genes using quantitative real-time (qRT)-PCR was confirmed ( Figure 8B). To further elucidate the potential regulatory mechanism of GhGPAT12/25, the known and predicted interactions among down-regulated genes were investigated, and genes with high connectivity with others in the protein-protein interaction (PPI) network were selected for functional analysis. Such investigation showed that these genes were also involved in lipid biosynthetic, wax metabolic, and cutin metabolic processes (Figure 8C), which correlated well with the GO enrichment of all down-regulated genes. Moreover, genes contributing to the biosynthetic processes of several secondary metabolites (isoprenoid, terpenoid and pigment) were also enriched in this network ( Figure 8C). These results implied that the mutation of GhGPAT12/25 might affect the biosynthesis and transportation of lipidic components required for the formation of exine and anther cuticle.
Anther Cuticular Waxes Were Altered in ghgpat12/25 Anthers Due to the defects of anther cuticle and pollen exine (Figures 5, 6), as well as the down-regulation of wax synthesis genes in ghgpat12/25 (Figure 8B), it can be speculated that the wax enrichment in ghgpat12/25 anthers was abnormal. To test this hypothesis, the compositions and quantity of cuticle waxes of the wild type and the ghgpat12/25 anthers were determined by gas chromatography-mass spectrometry (GC-MS) and gas chromatography-flame ionization detection (GC-FID). After clarifying the ratio between surface area and weight of anthers in WT and ghgpat12/25 (Supplementary Figure 5), the wax compositions per unit area (mm 2 ) were quantified. Alkanes were determined as the major wax monomers in WT and ghgpat12/25 anthers ( Figure 9A). Approximately, a total of 0.058 µg waxes per mm 2 were detected in WT anthers. However, the waxes in ghgpat12/25 anthers (0.014 µg/mm 2 ) were 75% less than those in the WT anthers ( Figure 9A and Supplementary Table 2). The affected waxes in ghgpat12/25 anthers were attributed to the reduction of alkanes (C24, C25, C26, C27, C29, C30, C31, and C32) and fatty acids (C20, C24, and C26) (Figures 9B,C and Supplementary Table 2). Additionally, a significant decrease in the levels of other metabolites, like tocopherols, campesterols, and diterpenoids, was also observed in ghgpat12/25 anthers (Figure 9D and Supplementary Table 2). On the other hand, the contents

GhGPAT12/25 Were Activated by GhMYB80
According to the analysis of the promoter sequences (Supplementary Figure 7), it was predicted that GhGPAT12/25 may be regulated by MYB or/and bHLH because of the existence of their binding sites. To further identify the potential upstream regulatory genes, MYB and bHLH transcription factors, which showed similar expression patterns to those of GhGPAT12/25, were screened based on the previous RNA-seq data of anther samples and other vegetative tissues (Zhang et al., 2015(Zhang et al., , 2020. Five pairs of homologous genes were found to be strictly expressed in anthers at the early developmental stages ( Figure 10A). Subsequently, a yeast one-hybrid system (Y1H) was performed using two sequences that represent the most likely binding sites of MYB or/and bHLH TFs (Figure 10B). For the TFs, the study revealed that only GhMYB80 (GhMYB80-A, GH_A04G0015) could directly bind to the WAACCA (MYB1AT) site in GhGPAT12/25 promoters ( Figure 10C). However, the other genes exhibited no or low binding activity on the SD/-Trp/-Leu/-His + 200mM 3-AT selective medium (Figure 10C).
GhMYB80 has been reported to share conserved sequences with that of Arabidopsis and Brassica and to rescue the male sterile phenotype of atmyb80 in Arabidopsis (Xu et al., 2014). The qRT-PCR analysis was carried out to identify the expression pattern of GhMYB80s in various tissues; thus, the specific expression characteristic in anthers during the TTP stage was determined ( Figure 10D). To explore their potential roles in cotton anther development and verify the regulatory relationship with GhGPAT12/25, a VIGS assay was used in TM-1 to transiently suppress the expression of GhMYB80s. A 300-bp fragment of the GhMYB80-A CDS from the nonconserved region was used in the (pCLCrVA)-based VIGS construct. As expected, pCLCrVA-GhMYB80s plants showed a significant decline in male fertility, characterized by 64%-93% of the anthers being shriveled (Figures 10E,F). The expressions of GhMYB80s in TTP stage anthers of the positive plants were down-regulated by approximately 59%-85% compared to the negative control plants (Figure 10G), implying that the corresponding phenotype was indeed due to the silence of GhMYB80s. Furthermore, the expression levels of GhGPAT12/25 in GhMYB80s-slienced plants were found to be significantly reduced in positive anthers ( Figure 10H). This suggested that GhGPAT12/25 act downstream of GhMYB80s and might be activated by GhMYB80s.
The activation of the GhGPAT12/25 promoters by the GhMYB80s protein was further confirmed via dual-luciferase (LUC) assays. In this system, the promoter sequence of GhGPAT12 was constructed to the pGreenII0800-LUC reporter plasmid, and the CDS of GhMYB80-A (GH_A04G0015) was cloned into an effecter plasmid (pGreenII62-SK) and driven by a CaMV35S promoter. The GhMYB80 transactivation level to the probable downstream promoter was measured by normalizing the signal of the luciferase reporter to that of Renilla luciferase (REN). The assay showed that the cotransformed effector-reporter exhibited a statistically significant higher-level transactivation in comparison with negative control (Figure 10I). This result indicated that GhGPAT12 can be activated by GhMYB80-A. According to the high similarity of the GhGPAT12/25 promoters (Supplementary Figure 7), it can be concluded that GhMYB80s are likely to directly bind to the specific elements in the GhGPAT12/25 promoters and

GhGPAT12/25 Are Required for the Anther Cuticle Formation and Pollen Exine Assembly
Heterosis utilization is an expectable goal in cotton genetic breeding. Male sterile lines are considered of great agricultural importance for the production of hybrids in order to improve the productivity and quality of cotton fiber (Zhu et al., 2008b). Few genes have been identified to be crucial for the formation of male organs in cotton (Wang and Li, 2009;Li et al., , 2020Liu et al., 2019a;Wang et al., 2019), and, therefore, the underlying regulatory mechanism controlling anther development and male sterility in cotton remains elusive.
The cytological observation and biochemical analysis of this research revealed that aberrant lipidic metabolism in ghgpat12/25 anther affects the precise tapetum degeneration and the biosynthesis of essential precursors in anther cuticle and pollen exine (Figures 4-7, 9 and Supplementary Figure 6). The tapetal layer of ghgpat12/25 could not undergo timely PCD, but displayed a series of unpredictable changes, such as lightly staining and swollen appearance, from the MC stage to the BNP stage (Supplementary Figures 6, 7). These abnormal morphologies are reminiscent of the phenotype of many male sterile lines, such as atgpat1 (Zheng et al., 2003) and lap5lap6 (Kim et al., 2010) in Arabidopsis; dpw , dpw3 (Mondol et al., 2019), and cyp704b2  in rice; and zmms33 (Zhang et al., 2018;Zhu et al., 2019) in maize.
Additionally, during the late developmental stage, the acute and abnormal degradative tapetum in ghgpat12/25 anther and the aborted pollen grains leaving the atrophic anther locules (Figures 4J,L) indicate defects in synthesis and transport of lipidic precursors from the tapetum cell to the pollen wall surface in the mutant. These defects confirm the essential role of the cooperative action between tapetum cell and microspore for pollen development.
The anormogenesis of microspores is another characteristic in ghgpat12/25. Microspores in ghgpat12/25 exhibited an irregular shape and abnormal exine (Figures 4-6). According to SEM detection, the pollen grains of ghgpat12/25 were relatively smooth with no reticulate exine pattern and spines on the pollen surface (Figures 5G,H). These deficiencies were caused by the poor development of sexine (tectum and bacula) on the pollen wall (Figures 5G,H). Such development was exposed by the TEM analysis (Figures 6J,P,V). Altogether, these deficiencies in tapetal layer PCD and microsporogenesis lead to the male sterility, and these results may be predicted by the expression of GhGPAT12/25 in both tapetal cells and pollen surface.
Despite the un-expression of GhGPAT12/25 on anther epidermal cells shown by the RNA in situ hybridization experiment (Figure 1), GhGPAT12/25 were also found to be critical for anther cuticle development (Figure 5). Anther cuticle is the protective barrier for the normal development of anthers and pollen grains in the anther locules (Piffanelli et al., 1998). As the downstream executor of anther lipid metabolism, GPAT is thought to be a committed enzyme for glycerol synthesis and an essential factor in the precise assembly of cuticle monomers (Li-Beisson et al., 2013). In the present study, SEM was performed to observe the anther surface of ghgpat12/25, thus discovering the seemingly lacking cuticle on anther surface (Figures 5E,F). This defect is consistent with the observations in osgpat3 (Men et al., 2017) and zmms33 (Zhang et al., 2018;Zhu et al., 2019) mutants. Moreover, this defect may be explained by the biochemical evidence establishing that the synthesis of wax monomers, such as alkanes and fatty acids, as well as several other metabolites (tocopherols, campesterols, and diterpenoids), was largely reduced in ghgpat12/25 anthers (Figure 9).
Therefore, these series of concomitant defects in pollen exine and anther cuticle highlight that the sporopollenin and cuticle may share similar lipidic constituents supplied by tapetal layer cells, and GhGPAT12/25 may act as important factors in affecting the normal functioning of tapetum during anther development.

GhGPAT12/25 Have Conserved and Diversified Functions for Anther Development in Dicot
The sn-Glycerol-3-phosphate O-acyltransferase (GPAT, EC 2.3.1.15) is the first enzyme for the assembly of membrane and storage glycerolipids (Zheng et al., 2003). Previous studies revealed that GPATs can be distinguished by the positions (sn-1 or sn-2) that catalyze the transfer of an acyl group from acyl-CoA/acyl-ACP to glycerol-3-phosphate (G3P) and whether they have phosphorylation activity. These characteristics have been extensively illustrated in animals and plants . In general, different structural characteristics and catalytic performance determine the diverse functions of GPATs. The sn-1 acyltransferases were found to play important roles in the production of intracellular storage oil and to be represented by ATS1 (Nishida et al., 1993) and AtGPAT9 (Gidda et al., 2009) in Arabidopsis. The sn-2 acyltransferases are likely involved in the biosynthesis of extracellular lipid barrier polyesters, cutin and suberin, such as AtGPAT1-8 , BnGPAT4 (Chen et al., 2014), OsGPAT3 (Men et al., 2017), ZmMS33 (Zhang et al., 2018;Zhu et al., 2019), and SlGPAT6 (Petit et al., 2016). Despite the scarce information about the molecular structure and biosynthetic steps of lipid polymers, the synthesis of cutin glycerolipid is produced via a variant set of reactions that store lipid biosynthesis (Yang et al., 2010).
Based on the protein conserved domain analysis, GhGPAT12/25 were identified as glycerol-3-phosphate acyltransferases (Figure 2). Phylogenetic tree analysis implied that GhGPAT12/25 were assigned to clade I and considered to be sn-2 acyltransferases homologous to AtGPAT1, indicating the potential roles of GhGPAT12/25 in polyester synthesis ( Figure 2B). Sequence alignment showed that GhGPAT12/25 and homologous members in other species share common conserved boxes (AT-I to AT-IV) in acyltransferase (AT) domain, with the corresponding catalytic and binding residues in all sequences ( Figure 2C). In addition, the HAD-like domain located in the N-terminal region of GPATs is thought to be a typical reference for distinguishing the existence of phosphohydrolytic activity. Like AtGPAT1/2/3/5/7, GhGPAT12/25 may not have this activity due to the absence of pivotal sites necessary for substrates binding and catalysis in DXD and GDXXXD motif, and they were predicted to acylate the sn-2 position of G3P and produce lysophosphatidic acid (LPA) (Figure 2C). In clade III, AtGPAT4, AtGPAT6, and AtGPAT8 contained intact activity sites in these motifs ( Figure 2C) and functioned as unique bifunctional enzymes with both acyltransferase and phosphatase activity, producing sn-2 MAC (monoacylglycerol) rather than LPA .
OsGPAT3 and ZmMS33 have been reported to be indispensable for male reproduction (Men et al., 2017;Zhang et al., 2018;Zhu et al., 2019), but the functions of their homologous genes in Arabidopsis (AtGPAT2/3) are not yet clear . This evidenced the functional differentiation of GPAT in dicotyledon and monocotyledon. Notably, AtGPAT1/6 are considered as crucial factors to anther and tapetum development in Arabidopsis (Zheng et al., 2003;Li et al., 2012). Similar to its ortholog AtGPAT1 in Arabidopsis, GhGPAT12/25 were also found to benefit male fertility in cotton, indicating their relatively conserved functions in dicotyledon. However, unlike AtGPAT1, which expresses in flower buds and siliques, causing a partial male sterility with defective pollen grains and normal anther cuticle with loss of function (Zheng et al., 2003), GhGPAT12/25 are specifically expressed in the early developmental anthers (Figure 1) and their mutants are completely male sterile with abnormal tapetum, immature exine, and a lack of cuticle (Figures 3-7), suggesting the diversified functions of GPAT during evolution. Altogether, these results reflect the conserved and diversified function of GhGPAT12/25 in male reproductive development. However, it is worth noting that more research on GPAT is needed to explore their specific evolutionary mechanism in dicots and monocots.

A Potential Regulatory Pathway of GhGPAT12/25 in Male Sterility
As an evolutionarily conserved gene in various plants, MYB80 is located downstream of AMS to control the anther development and cuticle formation. MYB80 is considered to directly or indirectly regulate the expression of a series of functional genes, which contribute to the synthesis or transport of lipidic components required for anther development, such as CYP703A2, MS2, LAP5, and LAP6 (Phan et al., 2011;Xiong et al., 2016;Lu et al., 2020). In addition, further genetic and biochemistry analysis have revealed that MYB80 directly activates the expression of MS1 and CYP703A2 by binding to the core motifs (AACC) located on the promoters Lu et al., 2020).
In this study, GhGPAT12/25 were found to be key enzymes in the synthetic pathway of lipidic monomers required for the formation of anther cuticle and pollen exine (Figures 5, 6). Their upstream regulator was then predicted to be GhMYB80s because of the similar expression patterns of them and the MYB1AT motif (CTAACCA) located on the promoter of GhGPAT12/25 ( Figure 10B). Y1H, dual luciferase, and VIGS assays, which were performed to investigate the potential regulatory mechanism of GhGPAT12/25 in anther development, demonstrated that GhMYB80s are likely to directly bind to the core motif of GhGPAT12/25 promoter, thereby activating the expression of GhGPAT12/25 (Figures 10C,E-I). These results preliminarily reveal the regulatory mechanism of GhGPAT12/25 in the process of anther cuticle formation and highlight the evolutionary conservation role of MYB80 in cotton anther development.
Indeed, except for the effects of MYB80 on anther cuticle formation, many MYB TFs, such as MYB16, MYB30, MYB94, MYB96, etc., were found to be crucial manipulators for the regulation process of cuticle synthesis in vegetative organs by regulating the expression of downstream lipidic-related genes (Oshima and Mitsuda, 2013;Lee et al., 2016;Zhang et al., 2019). Moreover, these downstream genes are mainly CER1, LACS2, KCS1, KCS2, KCS6, KCR1, and CER3, which are also involved in the formation of cutin and wax in anthers, thus implying that these genes may be regulated by diverse MYBs in different synthesis pathways of cuticle due to their ubiquitous expression in various tissues (Seo et al., 2011;Lee et al., 2016). Therefore, as key enzymes in anther cuticle formation pathway, GhGPAT12/25 may also be regulated by MYBs. Altogether, these discussions ratify the underlying regulatory mechanism of GhGPAT12/25 discovered in this research.
In view of our transcriptome analysis showing the significantly differential down-regulation of a large number of lipid biosynthetic and glycerol transport genes (SNH1, CER1, CER2, CER3, CER6, LACS1, LTPG1, LTPG6, ABCG10, ABCG14, and ABCG25, etc.) in ghgpat12/25 anthers ( Figure 8B and Supplementary Table 1), it can be speculated that GhGPAT12/25 likely function in modulating the synthesis and transport of lipidic framework during anther development, while their loss of function leads to the disturbances of these processes.
In previous researches, SNH1, CER1, CER2, CER3, and CER6 are considered to contribute to the production of epicuticular wax (Millar et al., 1999;Fiebig et al., 2000;Ariizumi et al., 2003;Broun et al., 2004;Haslam et al., 2012) and not affect the content and composition of cutin. LACS1 is an acyl-CoA synthetase that acts on long-chain and very-long-chain fatty acids, and it is important for the biosynthesis and assembly of both cuticular wax and cutin polyester in Arabidopsis (Lu et al., 2009). Therefore, according to the potential role of GhGPAT12/25 for the formation of cutin monomers, we speculated that the down-regulation of these wax-related genes may be due to the lack of cutin matrix that supports the correct incorporation of intracuticular and epicuticular waxes into the cuticle backbone. In other words, the failed cutin synthesis process affected the production of wax components. This hypothesis was confirmed by the reduction of wax components in ghgpat12/25 anthers (Figure 9).
Moreover, the lipidic monomers synthesized in tapetal cells need to be transported to the anther epidermis. This process requires the participation of ABCG transporters for the channeling of intracellular lipids through the plasma membrane and LTP(G)s responsible for exporting lipidic monomers to pass through the apoplastic compartment or hydrophilic cell wall (Edstam and Edqvist, 2014). Several ABCGs and LTP(G)s were found to be down-regulated in ghgpat12/25 anthers ( Figure 8B and Supplementary  Table 1); among them, LTPG1 has been reported to bind to lipids and to function as a component of the cuticular lipid export machinery in Arabidopsis. The mutation of LTPG1 causes the alterative compositions of cuticular lipid on epidermis (Debono et al., 2009). Meanwhile, LTPG6, which is abundantly expressed in pollen, has previously been found to affect the pollen morphology, ovule fertility, and seed formation (Edstam and Edqvist, 2014). LTPG26 is considered a young anther-specific gene, but its function is still unclear. Additionally, several genes, such as DRN1 (Dhar et al., 2020), LTP4 (Deeken et al., 2016), ABCG14 (Wang et al., 2017), and ABCG25 (Kuromori et al., 2010), were reported to be responsible for systemic resistance, especially for bacterial/fungus pathogens. It is worth considering that, although there is not enough evidence, these genes may exert resistance by transporting lipidic components used to synthesize cuticle. Other genes are mainly functionally ambiguous in Arabidopsis because of their unique effect characteristics. For example, ABCGs are half-transporters and require dimerization to function. Different dimer combinations may have diverse substrates, thus performing different functions (McFarlane et al., 2010). Therefore, their down-regulation in ghgpat12/25 may imply their potential roles in anther cuticle development, and it is indispensable to explore the detailed biological functions of ABCGs and LTP(G)s members in future studies, especially their effects in anther development.
In conclusion, a pair of paralogs GhGPAT12/25 specifically expressed in early developmental stage tapetum and microspores were identified in this research. CRISPR/Cas9mediated knockout and gene regulatory mechanism analysis demonstrated that GhGPAT12/25 are indispensable enzymes acting downstream of GhMYB80s and contributing to the anther cuticle formation and pollen exine development by synthesizing glycerol framework. Moreover, the loss of function of GhGPAT12/25 affects the expression of genes related to the synthesis and transport of cutin and wax, thus leading to the male sterile phenotype in ghgpat12/25. This study discovered pivotal genes controlling male sterility and provided important insights into the regulatory mechanism underlying anther development in cotton.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author/s.