An L-type lectin receptor-like kinase promotes starch accumulation during rice pollen maturation

The development of male gametophytes is one of the key steps in the plant reproductive process (McCormick, 2004). The development of pollen, the male gametophyte of flowering plants, starts from the differentiation of sporogenous cells and involves a series of complex events: pollen mother cells (PMCs), which are generated from sporogenous cells, undergo meiotic division (Ma, 2005). Subsequently, the individual haploid microspore with surrounding primexine is freed from the tetrads and then vacuolates into a round shape (Borg et al., 2009). Soon thereafter, with the gradual disappearance of the vacuole and the complete formation of pollen wall (consisting of the outer exine and inner intine), the microspores go through two-step mitotic division. During this stage, the accumulation of pollen starch initiates. After the second mitotic division, microspores develop into mature pollen grains (Zhang et al., 2011).

Pollen development requires a sufficient source of energy and carbohydrates (McCormick, 1991). Starch is one of the most conventional energy substances and carbohydrate source in the development of pollen; and starch is also the main storage material in the cytoplasm of mature pollen (Baker and Baker, 1979). During the last stages of pollen development, a rapid phase of starch accumulation occurring inside the microspore is crucial for pollen maturation and viability (Datta et al., 2001). This final phase not only provides the reserve of energy for pollen germination but also establishes pollen maturity, especially in many monocots (Datta et al., 2003).

The sequential starch biosynthetic pathway is well documented in the development of heterotrophic seeds (Okita, 1992; Tetlow et al., 2004). For example, sucrose is unloaded from the phloem termini into basal endosperm cells by plasma membrane-localized sucrose transporters and by SUGARS WILL EVENTUALLY BE EXPORTED TRANSPORTERs, or hydrolyzed first by cell wall invertase (INV) (Riesmeier et al., 1994; Eom et al., 2015). Subsequent cleavage of sucrose in starch biosynthesis is mediated by sucrose synthase (SUS) and INVs (Tetlow and Emes, 2019; Braun et al., 2014). SUS hydrolyzes UDP and sucrose to UDP-glucose (UDP-Glc) and fructose (Karrer and Rodriguez, 1992), with UDP-Glc being further converted to glucose-1-phosphate (Glc-1-P) by UDP-Glc pyrophosphorylase (UGP) (Long et al., 2017). INV converts sucrose to fructose and glucose (Chourey et al., 2006), which are further converted into Glc-1-P through a series of enzyme-catalyzed reactions (Kumar et al., 2018). The resulting Glc-1-P is then catalyzed by ADP-glucose (ADP-Glc) pyrophosphorylase (AGP) to form ADP-Glc (Lee et al., 2007), which serves as the glucosyl unit donor for the growing starch polymers (Okita, 1992; Börnke and Sonnewald, 2011).

Similar to seeds, pollen grains are also heterotrophic organs, which are symplastically isolated from the surrounding somatic cell layers. Unlike the sporophytically controlled exine formation, it has long been suggested that intine development and pollen starch accumulation are mainly contributed by gametophytic parts (Shi et al., 2015; Lee et al., 2016). However, few genes involved in starch metabolism of developing pollen have been identified and the mechanism by which pollen starch accumulates remains largely unknown. The RNA-interference (RNAi) plants of OsINV4 show disruption in pollen starch formation (Oliver et al., 2005). OsUGP2 mRNA is spatially and temporally found in binucleate microspores and mature pollen (Mu et al., 2009). Downregulation of this gene results in failed accumulation of starch in pollen, thus leading to male sterility (Mu et al., 2009). RA68 is a gene expressed preferentially in shoots and flowers with unknown function; RA68 knockdown plants show a reduction in pollen viability due to severely defective starch accumulation (Li et al., 2010). A homozygous mutation of the rice hexokinase 5 gene (OsHXK5) leads to significantly reduced pollen starch content (Lee et al., 2019). Phosphoglucomutase (PGM) catalyzes the reversible conversion between Glc-1-P and glucose-6-phosphate (Esposito et al., 1999). The expression of the rice plastidic PGM (OspPGM) is upregulated in maturing pollen when starch accumulation begins, and the pollen grains of the osppgm mutant displayed a starch-null phenotype (Lee et al., 2016).

The L-LecRLKs are a subfamily of RLKs with an intracellular serine/threonine kinase domain, a transmembrane domain and an extracellular legume-like lectin domain (Barre et al., 2002). Recent studies have suggested that L-LecRLKs are involved in plant-pathogen interactions, the stress response and plant development (Vaid et al., 2013; Bellande et al., 2017). In this study, we discovered a novel role for AP1, an L-LecRLK, in starch accumulation during pollen maturation. A mutation in AP1 caused decreased pollen starch accumulation, altered expressions of numerous genes and downregulated phosphorylation of several proteins involved in the starch and sucrose metabolism pathway in maturing pollen. In addition, our data suggest that AP1 directly interacts with and phosphorylates OsUGP2 to increase its enzymatic activity for pollen starch accumulation. Our work thus enhances the current understanding of the molecular basis of the L-LecRLKs functions and rice pollen starch metabolism.

By screening the mutant library, we identified a male-sterile mutant, namely ap1. During the vegetative stages of development, the ap1 mutant showed no obvious difference from the wild-type 9311 (an indica rice cultivar; WT) (Fig. 1A). At the heading stage, the flower morphology of the WT and ap1 were comparable (Fig. 1B,C); however, the mutant had slightly smaller and pale-yellow anthers compared with that of WT (Fig. 1D). By fluorescein diacetate (FDA) staining, we found that the ap1 pollen grains had an abnormal shape and could not be stained (Fig. 1E-H), indicating that the mutant pollen grains were not viable. I₂-KI staining indicated that the mutant pollen grains contained no starch (Fig. 1I,J). Moreover, viewing the pollen grains under high magnification showed that the mutant pollen grains were almost hollow and lacked typical inclusions (Fig. 1K,L). These findings indicated that the ap1 mutant is male sterile.

For the genetic analysis, we pollinated ap1 with WT pollen grains. The resulting F₁ plants were fertile; and the derived F₂ progeny displayed a nearly 3:1 ratio of fertile and sterile plants (Table S1), indicating that the male-sterile phenotype of ap1 is controlled by a single recessive locus. Next, we used the MutMap method to map this locus (Abe et al., 2012). By analyzing the re-sequencing data of the mutant, we found a candidate region on the long arm of chromosome 2 harboring a cluster of single nucleotide polymorphisms (SNPs) with index of 1 (Fig. 2A, Fig. S1, Table S2). A SNP occurred in LOC_Os02g26160 (C1759T) that introduced a stop codon (Fig. 2B), which caused premature translational termination after the 587th residue (Fig. S2). Our further genotypic analysis of the F₂ individuals showed that all the male-sterile plants (Fig. 2C, bottom) had this homozygous mutation (Fig. 2C, top), while the plants with normal fertility (Fig. 2C, bottom) were heterozygous or homozygous for the WT genotype (Fig. 2C top). Besides, this mutation was also segregated with a ratio of 1:2:1 in the F₂ population (Table S1). These results suggested that LOC_Os02g26160 might be the causal gene of ap1.

LOC_Os02g26160 encodes a putative L-LecRLK (Vaid et al., 2012), which contains a N-terminal signal peptide, a Lectin_legB domain (Pfam: PF00139), a transmembrane region and a protein kinase (PK) domain (Pfam: PF00069) (Fig. 2D). Modeling of protein tertiary structure indicated that the ap1 mutation resulted in the absence of three α-helixes of the PK domain and the C terminus (Fig. 2E), suggesting this mutation might impair protein function. Phylogenetic analysis showed that, although this L-LecRLK protein is more closely related to its monocot orthologs, it constitutes an isolated branch (Fig. 2F, Fig. S3), suggesting a potential functional diversification between this L-LecRLK protein and its homologs in other plant species.

To conduct a functional analysis, we designed two independent targets (Fig. 2B,D) and used the CRISPR/Cas9 method to knockout LOC_Os02g26160 in Nipponbare (a japonica rice cultivar) plants. Sequencing analysis identified five CRISPR/Cas9-mediated (cr) mutants with different types of homozygous mutations (Fig. 3A). An analysis of amino acids showed that all these mutations led to frame shifts, which in turn caused the pre-termination of protein translation (Fig. 3A, Fig. S4), suggesting the disruptions of LOC_Os02g26160 function within these mutants. Notably, two cr mutant lines (cr-2-3 and cr-2-11) (Fig. 3A) generated from editing of Target 2 lacked the C-terminal PK domain (Fig. 3B), similar to that of ap1. The following phenotypic observations revealed that, compared with Nipponbare, all of these cr mutants had yellow-white anthers (Fig. 3C) and produced hollow pollen grains (Fig. 3D) but displayed normal vegetative growth and flower morphology (Fig. S5A,B), mimicking the phenotypes of ap1.

Additionally, we performed a functional complementation analysis by transforming a Nipponbare genomic fragment – which included a ∼2 kb promoter region, 2 kb coding region and 0.8 kb 3′ UTR of LOC_Os02g26160 – into the ap1 and cr mutants. The positive transgenic plants all showed rescue of the male-sterile phenotype of both the ap1 (Fig. 3E,F) and cr (Fig. 3G,H) homozygous plants. These results confirmed that the male-sterile phenotype in the mutants is indeed caused by dysfunction of LOC_Os02g26160/AP1, indicating a crucial role of AP1 in the development of male gametes.

Quantitative real-time PCR (qRT-PCR) analysis revealed that AP1 was mainly expressed in spikelets with developing anthers and roots (Fig. 4A). Interestingly, AP1 expression gradually increased with anther development and reached its first peak in spikelets with anthers at stage 8 or 9, and then decreased but rapidly increased again to its maximum level in spikelets at stage 11 or 12. GUS activity in transgenic plants with the GUS reporter gene driven by the native promoter of AP1 also showed this expression pattern (Fig. 4B). Importantly, GUS staining also revealed a high accumulation of AP1 transcripts in the anthers at stage 11 (Fig. 4C), especially in microspores (Fig. 4D), which was further confirmed by RNA in situ hybridization (Fig. 4E,F).

Using the WoLF PSORT program (https://wolfpsort.hgc.jp/), AP1 was predicted as a plasma membrane-localized protein. The C-terminal PK domain of AP1 is likely intracellularly located (Fig. S6) (TMHMM 2.0 program, http://www.cbs.dtu.dk/services/TMHMM/). As expected, transient expression of an AP1-YFP fusion protein construct (Fig. 4G) in rice protoplasts (Fig. 4H-L) and Nicotiana benthamiana (tobacco) leaf epidermal cells (Fig. 4M-P) showed that AP1-YFP was well overlapped with the signals of a plasma membrane marker (OsPIP2.1-mCherry) (Dangol et al., 2017), indicating this protein was exclusively targeted to the plasma membrane. These results established that AP1 is a plasma membrane protein that is primarily located in microspores.

To detect possible cellular defects of anther and pollen development in ap1, we examined the chromosome behaviors of WT and ap1 PMCs during the meiotic process by 4′,6-diamidino-2-phenylindole (DAPI) staining. The results showed that the mutant meiotic cells had a WT-like chromosome synapsis and condensation, as well as cytokinesis (Fig. S7A), suggesting that the mutation of AP1 might not affect the meiotic process. Using Aniline Blue staining, we also observed similar callose signals of tetrads in the WT and the mutant (Fig. S7B).

We compared the cross-sections of developing anthers in the WT and ap1 according to the stages divided by Zhang and Wilson (2009). From stages 8a to 9, PMCs of both the WT and the mutant underwent meiosis normally (Fig. 5A,B,E,F), and then released free haploid microspores from tetrads (Fig. 5C,G), in agreement with our previous observations. Moreover, no significant defects were detected in the mutant anthers until stage 11 (Fig. 5D,H,I,M). At late stage 11, the mutant microspores appeared to have less cytoplasm content (Fig. 5N) compared with that of WT (Fig. 5J). Subsequently, the mutant microspores showed a severely impaired accumulation of storage material (Fig. 5O), which resulted in abnormally hollow and deflated pollen grains in the anther locule (Fig. 5P). In contrast, the WT microspores became maturing and then engorged with cytoplasm and starch granules (Fig. 5K,L).

We used electron microscopy to further understand the ultrastructural defects in ap1. The transmission electron microscope (TEM) results revealed that the mutant microspores vacuolated normally and thereafter formed a sickle shape with the complete deposition of the double-layered exine during stage 10 to early stage 11 (Fig. 6E-H), resembling the WT (Fig. 6A-D). At late stage 11, the WT microspores had a normal pollen wall, consisting of the exine and intine layers, and began to accumulate starch in the cytoplasm (Fig. 6I-K). However, the mutant microspores neither formed any intine-like structures nor showed signs of inclusion storage (Fig. 6L-N). At stage 12, the spherical WT pollen grains were fully packed with inclusions, such as starch granules (Fig. 6O-Q). In contrast, the cytoplasm of the mutant microspores degraded further, leaving an empty cellular space inside a shrunken exine layer (Fig. 6S-U). These results were consistent with the light microscope observations.

In addition, we found that, although the inner Ubisch bodies and the outer cuticle of the mutant anther wall (Fig. 6U,V) appeared to have developed normally, the arrangements of these two structures were denser than that of the WT (Fig. 6Q,R). Furthermore, we found an abnormally flat aperture in the mutant microspores (Fig. 6T) compared with the thick annulus surrounding the aperture of WT microspores (Fig. 6B). In support of these observations, further scanning electron microscope (SEM) analysis of the stage 12 anthers was carried out. The results showed that, compared with the WT (Fig. S8A-D,I-L), ap1 displayed thinner anthers (Fig. S8E) with more tightly distributed cuticular ridges (Fig. S8F,G) and Ubisch bodies (Fig. S8H), as well as pollen grains that lacked the annulus structure around the aperture (Fig. S8M-P). We also examined the cellular features of developing anthers and the surface of mature pollen in the cr mutants using cross-sections and SEM analysis, respectively. The results showed that the cr mutants also exhibited deficient engorgement of pollen storage materials (Fig. S9) and abnormal annulus of the pollen aperture (Fig. S10), which were identical to the defects observed in ap1. These findings demonstrated that dysfunction of AP1 not only caused the defective accumulation of pollen storage materials, but also affected the formation of the intine and aperture during pollen development.

To further explore the roles of AP1 in the late stages of pollen development, a transcriptome analysis of spikelets with anthers during stages 11 to 12 from the WT and ap1 plants was conducted. RNA-seq data showed that the ap1 mutation resulted in thousands of differentially expressed genes (DEGs), including 1221 downregulated and 936 upregulated transcripts (Fig. 7A). A Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis revealed that these DEGs were assigned into several functional classes. Among these classes, the biological processes of starch and sucrose metabolism were significantly enriched (Fig. 7B). In this class, a number of genes putatively related to sugar signaling and conversion, such as LOC_Os01g69030 (sucrose-phosphate synthase, SPS), LOC_Os06g14510 (sugar isomerase, SIS), LOC_Os03g53790 (periplasmic β-glucosidase, PBGlu), OsHXK7 (cytosolic hexokinase) (Kim et al., 2016), Os9bglu30 (β-glucosidase) and LOC_Os06g46340 (glycosyl hydrolase, GH) were downregulated in ap1 (Fig. 7C). The identical reduction in transcript level of these genes in ap1 and the cr mutants was also validated by our further qRT-PCR data (Fig. 7D, Fig. S11). Considering the close connection between sugar metabolism and starch biosynthesis (Chourey et al., 1998; Braun et al., 2014), these results implied a disordered sugars metabolism that may result in the deficiency of starch biosynthesis in ap1.

Starch is one of the main components in pollen storage materials and is necessary for pollen viability (Datta et al., 2001). We therefore determined the starch content in anthers from the WT and ap1 plants during pollen maturation. In agreement with the observed defects of pollen inclusion storage in ap1 (Fig. 1I-L), the mutant showed significantly decreased levels of starch at the late pollen development stages compared with that of the WT (Fig. 7E). These results supported that AP1 may participate in pollen starch accumulation.

To verify the kinase activity of AP1, we fused the C-terminal PK domain of AP1 with glutathione S-transferase (GST; GST-AP1-C) and performed an in vitro kinase assay. The results showed that the GST-AP1-C incubated with phosphorylation buffer (PB) has a significantly stronger phosphorylation level than that of the unincubated GST-AP1-C (Fig. 8A, Fig. S12), supporting that AP1 has kinase activity and can be auto-phosphorylated.

To gain insight into the putative downstream members of AP1 in the accumulation of pollen storage materials, we used a quantitative phosphoproteomic approach to investigate its potential in vivo targets. We analyzed the anther samples from WT and ap1 plants during stages 11 to 12. This analysis identified that a total of 2246 unique phosphorylation sites from 1449 proteins showed >1.5-fold difference (P<0.05) between WT and ap1 (see Table S5), of which 940 phosphoproteins were downregulated and 509 proteins were upregulated in the mutant (Fig. 8B). As AP1 is an active kinase, we assumed that its targets would reduce the phosphorylation level in the mutant plants. We then focused on the downregulated phosphoproteins. The KEGG pathway analysis indicated that the downregulated phosphoproteins were enriched in serval pathways (Fig. 8C). Consistent with the defective starch accumulation in ap1 pollen grains, key enzymes involved in the starch and sucrose metabolic pathway, such as OsHXK5 in sugar cleavage, and OsUGP2 and PGMs in Glc-1-P production, were also significantly enriched (Fig. 8C and Table S3). Further investigation showed that the phosphorylation levels of OsUGP2 were apparently decreased at two serine residues (S151 and S413; Fig. 8D, Fig. S13, Table S3). To exclude the possibility that the downregulation of OsUGP2 phosphorylation is correlated with its mRNA transcript level, we quantified the expression level of OsUGP2 and detected no significant differences between the WT and ap1 mutant (Fig. S14A), suggesting this was a direct consequence of the loss of AP1 phosphorylation function. Similar to AP1 (Fig. 4A-F), OsUGP2 is highly expressed in the spikelets with anthers at stages 11 to 12 (Fig. S14B), and its protein is also predominately located in maturing pollen grains (Mu et al., 2009). Moreover, the phenotype of ap1 (Figs 1 and 5) is similar to that of the OsUGP2 knockdown plants (Mu et al., 2009). Considering these similarities and the important role of OsUGP2 in starch accumulation during rice pollen maturation (Mu et al., 2009), we speculated that OsUGP2 is a potential candidate for AP1 substrates.

Previous studies have reported that plant PKs typically phosphorylate their substrates by physical interactions to modulate developmental and biosynthetic processes (Mao et al., 2011; Tian et al., 2017). It is worth noting that, apart from the downregulated phosphorylation of OsUGP2 in the ap1 mutant, OsUGP2 and AP1 are co-expressed at the stage of pollen starch accumulation (Fig. 4A-F, Fig. S14B) (Mu et al., 2009). The subcellular localization assay also showed the overlapped distributions of co-expressed AP1-YFP and OsUGP2-CFP (Fig. S15). Therefore, we used a yeast two-hybrid (Y2H) assay to determine whether AP1 could interact with OsUGP2 directly. Indeed, the results of the Y2H assay revealed a positive interaction between the AP1-C and OsUGP2 in yeast cells (Fig. 9A).

To validate the interaction of AP1 with OsUGP2, we conducted the bimolecular fluorescent complementary (BiFC) assays in tobacco leaves. We found that both full-length AP1 protein and AP1-C were able to interact with OsUGP2 in tobacco leaf epidermal cells (Fig. 9B). In contrast, no positive interaction was detected between the AP1 mutated protein and OsUGP2 (Fig. 9B), suggesting that the PK domain of AP1 is responsible for mediating AP1-OsUGP2 interactions. Subsequently, we performed a firefly luciferase complementation imaging (LCI) assay in tobacco leaves (Chen et al., 2008). As expected, the co-expression of OsUGP2-CLuc with NLuc-AP1-C or NLuc-AP1 constructs in tobacco leaves also confirmed the interaction between AP1 and OsUGP2 (Fig. 9C). In addition, GST pull-down assays revealed that a complex containing the GST-AP1-C and the His-tagged OsUGP2 protein (His-OsUGP2) formed in vitro (Fig. 9D), whereas the His-OsUGP2 protein did not bind to the mutated PK domain of AP1 protein (GST-ap1-C) (Fig. 9D). Taken together, both in vivo and in vitro results indicated that AP1 directly interacts with OsUGP2 through its PK domain.

The interaction between AP1 and OsUGP2 suggests that OsUGP2 may be a direct phosphorylation target of AP1. In order to investigate whether OsUGP2 can be phosphorylated by AP1, recombinant His-OsUGP2 protein was incubated with GST-AP1-C or GST-ap1-C for a phosphorylation assay. The immunoblotting results showed that the PK domain of AP1 strongly trans-phosphorylated OsUGP2 in vitro, whereas the AP1 mutated version of the PK domain lost auto-phosphorylation activity and failed to phosphorylate OsUGP2 (Fig. 10A). In addition, changing both S151 and S413 to alanine (OsUGP2^AA) largely impaired the phosphorylation of OsUGP2 by GST-AP1-C (Fig. 10A).

To explore the effects of the phosphorylation of OsUGP2 by AP1, we measured whether the enzymatic activity of His-OsUGP2 changed with the presence of AP1 in vitro. As shown in Fig. 10B, when co-incubated with GST-AP1-C, His-OsUGP2 exhibited higher levels of enzymatic activity than did His-OsUGP2 alone and co-incubation of His-OsUGP2 and GST-ap1-C. To test whether the phosphorylation of S151 and S413 is required for OsUGP2 activity, we created another mutated version of OsUGP2, OsUGP2^DD (changing both S151 and S413 to aspartic acid), which mimics the phosphorylated S151 and S413. The results showed that, unlike His-OsUGP2, the enzymatic activity of His-OsUGP2^AA was lower than that of His-OsUGP2^DD, and could not be increased by incubation with GST-AP1-C (Fig. 10B).

To confirm the effects of the S151 and S413 phosphorylation of OsUGP2 in planta cells, we transiently expressed OsUGP2 and its variants in tobacco leaf cells and assessed their effects on starch content. The results showed that cells expressing OsUGP2^DD and cells co-expressing AP1 and OsUGP2 had a similar starch content, which was significantly higher than that of OsUGP2 alone and the combination of ap1 and OsUGP2 (Fig. 10C). Cells expression of OsUGP2^AA alone had a low starch content as the controls (Fig. 10C). Therefore, our in vitro and in vivo data suggested that AP1 directly affects OsUGP2 activity to promote starch biosynthesis possibly by post-translational modifications of phosphorylation.

Mature pollen grains of rice are usually surrounded by a pollen wall and are typically full of stored carbohydrates such as starch (Feng et al., 2001; Zhang et al., 2011). In rice, previous studies have revealed the factors that contribute to pollen wall formation, such as Acyl-CoA Synthetase12, Polyketide Synthases1/2 and Strictosidine Synthase-like2 for exine formation (Zou et al., 2017a,b, 2018), as well as Collapsed Abnormal Pollen1, ATP Binding Cassette G Transporter3 and Glycosyltransferase1 for intine formation (Moon et al., 2013; Ueda et al., 2013; Luo et al., 2019). However, the underlying mechanism by which storage materials accumulated during pollen maturation remains largely unknown. Here, we presented findings supporting the novel role of AP1, an L-LecRLK, in pollen starch accumulation during pollen maturation in rice.

Morphological observations suggest that loss-of-function of AP1 resulted in abnormal and nonviable pollen with little or no pollen storage materials (Fig. 1). According to our cytological observations (Figs 5 and 6, Fig. S8), the mutant pollen underwent normal meiosis, haploid microspore release and vacuolation, but showed subcellular abnormality during pollen maturation. The defected accumulation of pollen storage was observed at this stage in the mutant. Starch is the main component of pollen cytoplasm in major cereal crops, its accumulation is required for pollen maturation and it is responsible for pollen viability (Wen and Chase, 1999; Datta et al., 2001). Consistent with this, we observed a significantly lower starch content in the anthers of mutant plants at late developmental stages (Fig. 7E). In addition, the mutant pollen grains also had a deformed aperture and intine (Fig. 6, Fig. S8). It has previously been suggested that pollen starch synthesis and intine formation are mainly controlled by the gametophyte (Shi et al., 2015; Lee et al., 2016). Our results provide new evidence on the crucial role of L-LecRLKs in later gametophytic pollen development in rice.

The genetic analysis indicated that the male-sterile phenotype of ap1 appeared to be contributed by the sporophyte (Table S1). It is therefore reasonable to speculate that AP1 may play multiple roles both in sporophytic and gametophytic pollen development. Unexpectedly, we did not find apparently sporophytic defects in the pollen exine and anther wall layers of ap1 mutant, the classical cell layers for sporophytic abnormalities. It is possible that the sporophytic defects of ap1 cannot be seen, but are sufficient to alter fertility. Very recently, three other allelic mutants of AP1 were reported: s13283 (in the background of the indica cultivar Huanghuazhan), oslecrk5 (in the background of the japonica cultivar 02428) and defective in aperture formation1 (osdaf1; in the background of the japonica cultivar 9522) (Peng et al., 2019; Wang et al., 2020; Zhang et al., 2020). We carefully compared the cytological characteristics of these allelic mutants and found that they all showed the following identical features: shrinking pollen grains with gametophytic defects (less or no cellular content and absence of pollen intine layer) and sporophytic defects (abnormal aperture morphology). These cytological observations further support the conclusion that AP1 plays a role in pollen starch metabolism during pollen maturation.

However, other sporophytic cellular defects among these allelic mutants were highly variable. For example, the pollen exine was thickened in s13283 grains (Peng et al., 2019), while the ap1 (Fig. 6, Fig. S8) and osdaf1 mutants seemed to display a WT-like pollen exine (Zhang et al., 2020). oslecrk5 showed impaired formation of the callose layer (Wang et al., 2020), while the ap1 (Fig. S7) and osdaf1 mutants displayed no obvious defects in callose deposition during meiosis (Zhang et al., 2020). In ap1 and oslecrk5 tapetum, the Ubisch bodies appeared denser or smaller than in the WT (Fig. S8D,H) (Wang et al., 2020), while in s13283 and osdaf1, these abnormalities were not obvious (Peng et al., 2019; Zhang et al., 2020). In addition, most of these mutant-specific cellular defects appeared relatively weak. These sporophytically cellular defects might support the hypothesis that the sporophytically male-sterile phenotype of ap1 is caused by a potentially minor sporophytic defect, such as the denser arrangements of Ubisch bodies. The differently cellular defects of theses mutants can be explained, to some extent, by their different genetic backgrounds. For example, the qRT-PCR analysis revealed that the transcript level of AP1 reaches two peaks (at stage 8 and stage 11, respectively) during pollen development in indica cultivar 9311 (Fig. 4A) and Huanghuazhan (Peng et al., 2019), whereas only one peak (at stage 8) was revealed in japonica cultivar 02428 and 9522 (Wang et al., 2020; Zhang et al., 2020). In addition, these diverse expression patterns are generally in accordance with the temporal and spatial occurrence of the cellular defects of the corresponding mutants. Therefore, these data collectively support the multiple roles of AP1, both sporophytic and gametophytic, during rice pollen development.

Starch serves as both a sign of pollen maturity and is also a main component of pollen storage materials (Datta et al., 2001). As the source of carbon, sucrose could be metabolized and used for starch synthesis in pollen (Datta et al., 2003). UGP degrades UDP-glucose to produce Glc-1-P, which further serves as the carbon skeleton for starch synthesis (Coleman et al., 2006). Silencing of OsUPG2, a member of the rice UGP family, blocked the starch accumulation in pollen, resulting in male sterility (Mu et al., 2009). These findings suggest an essential role for OsUPG2 in pollen starch biosynthesis and pollen maturation. In the present study, the most important finding is that mutation of AP1 reduced the starch accumulation within pollen (Figs 1, 2, 5, 6 and 7). Two distinctly downregulated phosphorylation sites, located at the UGP domain of OsUPG2, were identified in the ap1 mutant (Fig. 8D, Fig. S13, Table S3), although the mutation did not cause apparent transcriptional changes of OsUGP2 (Fig. S14A). Remarkably, both AP1 and OsUGP2 are highly co-expressed during late pollen development (Fig. 4, Fig. S14B), which is similar to the phase of pollen starch biosynthesis (Mu et al., 2009). These findings suggest that dysfunction of AP1 might directly lead to the reduced phosphorylation levels of OsUPG2. Consistently, our data indicated that AP1 interacted with and phosphorylated OsUGP2 (Figs 9A-D, 10A), whereas the mutants did not (Figs 9B-D, 10A). Typically, the extracellular domain of RLKs binds the ligands, while the intracellular PK domain regulates the activity of targets through interaction and transphosphorylation (Walker, 1994). Therefore, as an active L-LecRLK, AP1 likely plays a role in pollen starch metabolism through phosphorylation of OsUGP2. Phosphorylated UGPs have been detected during starch accumulation in seeds (Zhang et al., 2014; Pang et al., 2018), suggesting that phosphorylation may be required for maintaining the normal enzymatic activity of OsUGP2 in pollen starch biosynthesis. Consistent with this suggestion, our in vitro and in vivo assays indicated that the phosphorylation of OsUGP2 by AP1 may enhance OsUGP2 activity (Fig. 10B,C). These findings support the hypothesis that AP1 targets OsUGP2 to exert its role in promoting pollen starch biosynthesis (Fig. S16).

The transcription of many genes associated with sugar and carbohydrate metabolism, as well as sugar signaling, was significantly altered in ap1. For example, the expression of a SPS-encoding gene (LOC_Os01g69030) was significantly decreased in ap1 (Fig. 7C,D). SPS proteins play a key role in the complex reaction network of sucrose and starch content. They catalyze a reversible reaction from UDP-Glc and D-fructose-6-phosphate to sucrose-6 phosphate, which is further dephosphorylated to sucrose by sucrose phosphatase (Huber and Israel, 1982; Huber, 1983; Chua et al., 2008). The resultant sucrose can be hydrolyzed to UDP-Glc by SUS for the downstream starch biosynthesis (Karrer and Rodriguez, 1992). As a potent sugar signal in plants, trehalose 6-phosphate (T6P) mediates sucrose use and starch metabolism (Lunn et al., 2006; Martins et al., 2013; Griffiths et al., 2016). T6P is synthesized via T6P synthase (TPS) but dephosphorylated to trehalose and Pi by T6P phosphatase (TPP) (Eastmond and Graham, 2003; Paul, 2007). We observed that the mutation of AP1 reduced the expression of two TPP genes and two TPS genes, whereas it upregulated the expression of one TPS gene (Fig. 7C,D), indicating an altered T6P homeostasis in ap1. HXKs are known sugar sensors that catalyze the conversion of glucose and fructose to hexose 6-phosphates, and play a role in sugar metabolism (Kim et al., 2016). Our data indicated that, in the mutants, OsHXK7 also showed reduced expression levels (Fig. 7C,D). The control of sucrose cleavage to monomers, sugar signaling and the downstream reactions affect the formation of starch (Kumar et al., 2018), suggesting the perturbed expression levels of these genes may also contribute to the deficient starch accumulation in ap1 pollen. Considering that AP1 mainly plays a role in post-translational phosphorylation modification, the abnormal expression of ap1 is likely to be a secondary consequence of its mutation.

Phosphorylation events are involved in the general regulation of starch biosynthesis (Pang et al., 2018). As an active PK, AP1 may modulate the functions of its targets by phosphorylation. The activity of starch synthesis enzymes may be affected by their phosphorylation level (Cohen, 1985). For example, in spinach, the change in site-specific phosphorylation of SPS was responsible for the activation state of SPS (Toroser et al., 1999). The enzymatic activity of PGM was enhanced by signaling kinase p21-activated kinase 1 through phosphorylation (Gururaj et al., 2004). Therefore, it is reasonable to propose that the downregulation of phosphorylation may have led to a reduced capacity for metabolic accumulation of sugar and starch, affecting the pollen viability during maturation. Our phosphoproteomic analysis identified downregulated phosphorylation levels of numerous proteins involved in the same pathway (Fig. 8, Table S3). For example, the phosphorylation levels of OspPGM and OsHXK5, which are crucial for pollen sugar metabolism and downstream starch biosynthesis (Lee et al., 2016, 2019), were downregulated in the ap1 mutant (Table S3). OspPGM and OsHXK5 are located in the plastid (Lee et al., 2016, 2019) (Table S3), suggesting that these two proteins may not directly interact with AP1. Consistently, we did not identify any interactions between AP1 and OspPGM in plant cells (Fig. S17), indicating that AP1 may also indirectly regulate downstream proteins to participate in pollen starch accumulation. In addition, the phosphorylation levels of other proteins that are putatively involved in sugar and starch metabolism, such as disproportionating enzymes (LOC_Os07g46790 and LOC_Os07g43390), SPSs (LOC_Os01g69030 and LOC_Os02g09170) and cytosolic PGM (LOC_Os03g50480), were also downregulated in ap1 mutant (Table S3). Sugar metabolism and starch biosynthesis require the concerted actions of an extensive range of genes and many sub-functional enzymes (Ball et al., 2011; Hwang et al., 2016), suggesting that, in addition to OsUGP2, AP1 may also affect the phosphorylation levels of other key enzymes, which possibly impact on pollen sugar and starch metabolism. Thus, it will be interesting to reveal how AP1 causes these changes on the expression or phosphorylation of key factors in sugar metabolism and downstream starch biosynthesis.

The ap1 plant was isolated from a library of ethyl methyl sulfonate-induced mutants that were generated in a background of the indica rice cultivar 9311 (Zou et al., 2017b). For genetic analysis and gene mapping, ap1 was back-crossed with the WT to produce the F₂ progenies. A japonica rice cultivar Nipponbare was used for knocking out AP1 via the CRISPR/Cas9 system and for the promoter-GUS assay. Rice plants used in this study were grown in the paddy field of Sichuan Agricultural University in Chengdu (Sichuan, China) or in Lingshui (Hainan, China) with routine management.

Plants, spikelets and anthers were photographed with a digital camera EOS-1200D (Canon). Pollen viability analysis was performed by staining with FDA and 1% I₂-KI solution, as described previously (Pinillos and Cuevas, 2008; Tao et al., 2019). For anther development analysis, the spikelets with anthers at differently developmental stages were specified according to Tao et al. (2019), and were fixed, dehydrated, embedded, sliced and stained subsequently as described previously (Zou et al., 2017b). The transverse section slides were photographed using an Axio Lab A1 microscope (Zeiss). The detection of chromosome behaviors was carried out according to Yi et al. (2012). For SEM and TEM analysis, anther samples were collected and processed as described by Zou et al. (2018) and then observed with a JSM-7500F SEM (JEOL) or JEM-1400PLUS TEM (JEOL), respectively. Starch measurements were performed as described previously (Zhang et al., 2013).

Leaves of 50 male-sterile mutants in the F₂ progenies were collected for DNA extraction. The extracted DNAs were then pooled and re-sequenced using the Illumina Hiseq platform (Novogene Bioinformatics Technology). Using the MutMap method (Abe et al., 2012), the sequencing results were computationally analyzed. Co-segregation analysis of the candidate mutation in F₂ population was carried out by PCR analysis and sequencing with primer set Seq-AP1. All the primers used in this study are listed in Table S4.

The full-length amino acid sequence of AP1 was used as a query for identifying its plant homologs through the Basic Local Alignment Search Tool in the National Center for Biotechnology Information database (https://blast.ncbi.nlm.nih.gov/Blast.cgi) with default parameters. The proteins with the highest identities to AP1 in each species were considered as the potential AP1 orthologs. Multiple protein alignments and phylogenetic tree constructions were performed using MEGA 5.0 software based on the ClustalW and neighbor-joining method (1000 bootstrap replications), respectively.

Specific targets for knocking out of AP1 by CRISPR/Cas9 system were designed using a CRISPR design tool (http://rice.hzau.edu.cn/cgi-bin/rice2/CRISPR_rice) from Rice Information GateWay. The following CRISPR/Cas9 constructions were performed using the methods described by Miao et al. (2013). A genomic fragment containing 1964 bp promoter region, 2088 bp coding region and 810 bp 3′-untranslated region (UTR) of AP1 was amplified from Nipponbare, and was thereafter cloned into the binary vector PHB to generate the AP1-com vector. Using the Agrobacterium-mediated transformation method with hygromycin B selection (Hiei et al., 1994), the resulting CRISPR/Cas9 constructs were introduced into Nipponbare for knocking out of AP1. For functional complementation analysis, the AP1-com vector was introduced into the callus induced by culturing of the young panicles of ap1 or cr homozygous (transgenic-free) plants with Glufosinate selection (Ling et al., 1983; Hiei et al., 1994).

A 1964 bp promoter region of AP1 was amplified and then cloned into the PHB-GUS vector to generate the AP1_pro::GUS construct, which was further introduced into Nipponbare by Agrobacterium-mediated transformation (Hiei et al., 1994). GUS staining was performed as described previously (Zou et al., 2018).

For RNA in situ hybridization, templates for the AP1-specific probes were generated using primer set Rish-AP1. The subsequent transcription, labeling and hybridization of probes were performed according to Zou et al. (2017a). Images were obtained using a Zeiss Axio Lab A1 microscope.

Total RNAs were extracted from the specified tissues using an RNeasy Plant Kit (Qiagen). RNA-seq was conducted by Novogene Bioinformatics Technology. More than 6 GB of data were yielded from each sample. Clean data were further analyzed according to Jeong et al. (2016).

The cDNA of extracted RNA was reverse transcribed using NovoScript Reverse Transcriptase (Novoprotein). qRT-PCR was performed on qTOWER3G machine (Analytik-Jena) using AceQ Universal SYBR qPCR Master Mix (Vazyme Biotech) according to the manufacturer's instructions. Relative expression levels were calculated following the 2^−ΔΔCt method and using the Ubiquitin (LOC_Os03g13170, Ubi) as the internal reference.

Coding sequences of the intracellular domains of AP1 and ap1 were amplified with primer set Gex-AP1-C and then cloned into pGEX-4T-1 for fusing in-frame with the GST tag. The point mutation constructs OsUGP2^AA and OsUGP2^DD were first generated with the Fast Mutagenesis System kit (Transgen) using specific primer pairs. OsUGP2 (including the mutated versions) full-length cDNAs were amplified and then cloned into pET28a for fusing in-frame with the 6×His tag. These constructs were introduced into Transetta (DE3) cells (Transgen). Isopropyl-β-D-thiogalactopyranoside (Transgen) was used for induction of protein expression. The recombinant proteins were purified using Mag-Beads GST Fusion Protein Purification kit (Sangon Biotech) or Ni-NTA-Sefinose Colum kit (Sangon Biotech). The in vitro phosphorylation assays were conducted according to the methods described previously with minor modifications (Zhang et al., 2017). In brief, the recombinant proteins were incubated in the PB [50 mM ATP, 50 mM Tris-HCl (pH 7.5), 10 mM MgCl₂, and 1 mM DTT] at 30°C for 2 h. The reaction was stopped by the addition of the 6×Protein Loading Buffer (Transgen). After that, the proteins with or without incubation were separated by SDS-PAGE. The phosphorylation was examined by immunoblotting analysis using anti-phospho (Ser/Thr) Phe antibody (Abcam, ab17464; 1:2000). The input proteins were immunoblotted using anti-GST Tag antibody (Sangon Biotech, D110271; 1:2000) or anti-6×His Tag antibody (Sangon Biotech, D191001; 1:2000).

Total protein was extracted from the anther samples of the WT and ap1 following the method described previously (Roy et al., 2007). After trypsin digestion, the resulting peptides were desalted and vacuum dried. The resulting peptides were labeled using the TMT Mass Tagging kit (ThermoFisher Scientific). The TMT-labeled peptides were fractionated into fractions by high pH reverse-phase HPLC. To enrich the phosphopeptides, peptide mixtures were incubated with Ti-IMAC microspheres, and were then collected, washed and eluted according to Tape et al. (2014). The enriched phosphopeptides were further collected and lyophilized, and were then subjected to LC-MS/MS analysis at PTM-BIO Biotechnology to identify the phosphorylation sites and levels.

Coding sequences of the intracellular domains of AP1 and the full-length OsUGP2 protein were amplified with primer sets AP1-C-BK and OsUGP2-AD, and then cloned into the Y2H bait pGBKT7 and prey pGADT7 vectors, respectively. The prey and bait constructs were co-transformed into Saccharomyces cerevisiae strain Y2H gold (Clontech) for performing the Y2H assay according to the manufacturer's instructions.

For subcellular localization, coding sequences (without the stop codon) of the full-length AP1 and OsUGP2 were amplified using the corresponding primer sets. The PCR products of AP1 and OsUGP2 were cloned into the binary vectors PHB-YFP and PHB-CFP for fusing with FP under control of the CaMV35S promoter, respectively. For BiFC assay, the full-length CDS of OsUGP2 was amplified and cloned into a binary vector pXY106 for fusing with N-terminal split nYFP (YN); meanwhile, the CDS encoding full-length AP1 and intracellular domains of AP1 (without the stop codon) was amplified and cloned into a binary vector pXY104 for fusing with C-terminal split-cYFP (YC), respectively. The resultant plasmids were transiently expressed or co-expressed in tobacco leaf epidermis cells or rice protoplasts following the methods described previously (Wang et al., 2013; Lichocka and Schmelzer, 2014). Fluorescent signals were observed using a Nikon A1 fluorescence confocal scanning microscope or Zeiss Axio Imager M2 fluorescence microscope.

The full-length AP1 and the intracellular domains of AP1 or ap1 were amplified and then cloned into the pCAMBIA1300-nLUC vector for fusing with N-terminal split luciferase (NLuc), respectively, whereas the full-length CDS of OsUGP2 was amplified and cloned into the pCAMBIA1300-nLUC vector for fusing with C-terminal split luciferase (CLuc). OsGIF1 and OsGRF4 proteins (Li et al., 2016) were used as the positive control. LCI assays were performed in tobacco leaves following the protocol described previously (Chen et al., 2008).

The recombinant GST-AP1-C or GST-ap1-C fusion proteins were immobilized on Glutathione Magarose Beads (Sangon Biotech) and incubated at room temperature for 30 min. The recombinant His-OsUGP2 protein was then added to the magnetic bead complex and incubated for another 30 min at room temperature. After three times of magnetic bead washing, the magnetic bead complex was subsequently eluted with elution buffer [50 mM Tris-HCl and 10 mM reduced glutathione (pH 8.0)]. The elution components were subjected to immunoblotting analysis using anti-GST tag and anti-His tag antibodies (Sangon Biotech, D110271 and D191001 at 1:2000, respectively).

The activity of UGPase was determined according to the protocol described previously (Woo et al., 2008) with slight modifications. Briefly, 0.2 μg of recombinant proteins were added to reaction mixtures (0.2 ml, pH 8) containing 0.2 µmol of UDP-glucose, 0.004 µmol of Glc-1,6-diP, 0.12 µmol of NADP, 1 U of phosphoglucomutase, 1 U of glucose-6-phosphate dehydrogenase, 4 µmol of Cys, 16 µmol of glycylglycine and 1 µmol of MgCl₂. Reactions were initiated with the addition of 0.5 µmol inorganic pyrophosphate. In this reaction, the UGPase converts UDP-Glc to Glc-1-P. Then, under the action of the phosphoglucomutase, G-1-P is used to generate glucose-6-phosphate. The resulting glucose-6-phosphate together with NADP are catalyzed by glucose-6-phosphate dehydrogenase to produce gluconolactone-6-phosphate and NADPH. The NADPH formation was monitored continuously at 340 nm and 30°C for 6 h with a molar extinction coefficient of 6.22×10³ M⁻¹ cm⁻¹ to reflect the enzyme activity of UGPase.

The full-length CDS sequences of OsUGP2 (including the mutated versions), AP1 and ap1 were amplified and then cloned into a binary vector PHB. The resultant plasmids were transiently expressed or co-expressed in tobacco leaf epidermis cells following the Agrobacterium infiltration methods described by Lichocka and Schmelzer (2014). After 48 h of infiltration, the corresponding tobacco leaf area was cut down and ground in liquid nitrogen. The starch content was then examined with the Starch Content Detection Kit (Solarbio).

We thank Dr Yangwen Qian (Hangzhou Biogle) for assistance with rice genome editing and transformation.