An atypical strictosidine synthase, OsSTRL2, plays key roles in anther development and pollen wall formation in rice

Strictosidine synthase (STR) plays an important role in the biosynthesis of terpenoid indole alkaloids (TIAs) and is expressed in a range of active meristematic tissues of higher plants. STR proteins are involved in different physiological and biochemical pathways. However, the function of STR proteins in rice development remains poorly understood. In this study, we identified 21 possible STR-like (OsSTRL) family members in rice genome and found that only one gene, OsSTRL2, exhibited a pre-emergency specific florescence expression pattern. Tissue-specific expression profile analysis, β-glucuronidase histochemical (GUS) staining and RNA in situ hybridization confirmed that OsSTRL2 was highly expressed in tapetal cells and microspores. Comparative protein sequence analysis indicated that OsSTRL2 lacked the key catalytic residue found in a typical STR (STR1), although it possessed conserved β-propellers and α-helices formed the basic structure of STR1. OsSTRL2 knockout mutant resulted to male sterility because of the defects in anther development and pollen wall formation. Subcellular localization of OsSTRL2-YFP revealed that the OsSTRL2 protein was primarily localized in the endoplasmic reticulum (ER). Therefore, OsSTRL2 is an atypical strictosidine synthase that plays crucial roles in regulating anther development and pollen wall formation in rice.


Results
Genome-wide identification of 21 OsSTRL genes in rice. BLASTP analysis was performed with STR1 protein sequence to identify the OsSTRL members in rice. Twenty-one putative OsSTRL protein sequences were obtained (Table 1) from the MSU Rice Genome Annotation Project Database (RGAP; http://rice.plantbiology. msu.edu/) 28 . To further confirm the reliability of these candidates, we performed conserved domain analysis by using the Simple Modular Architecture Research Tool (SMART) 29 . All OsSTRLs were detected within the "Str_synth" domain (PF03088.9) in their protein sequences. Through genomic distribution analysis, we found that the 21 OsSTRL members were distributed across chromosomes 1, 3, 6, 7, 8, 9, 10, 11, and 12. These genes were not found in the other three chromosomes (Table 1).
To understand the relationship among the OsSTRL genes, we used corresponding protein sequences from these genes and STR1 to perform peptide alignment by employing Clustal W (www.ebi.ac.uk/Clustalw). The result showed significant conservation within the strictosidine synthase domain among the OsSTRL proteins (Supplementary Figure 1). A neighbor-joining (NJ) phylogenetic tree was constructed by using the above multiple sequence alignment results with bootstrap analysis (1,000 replicates) (Supplementary Figure 2) to obtain clues about the evolutionary history of the OsSTRL genes. To analyze the conserved motifs of OsSTRL proteins, we employed the MEME motif search tool to investigate the shared motifs 30 . The OsSTRL genes were manually divided into two major groups based on the bootstrapping values of phylogram (Supplementary Figure 2) and the conserved motif distribution of OsSTRL proteins (Supplementary Figure 2). Group I is composed of four clades and three singletons, namely, OsSTRL19, OsSTRL3, and OsSTRL18 (Table 1 and Supplementary Figure 2); STR1 acted as a singleton in this group (Table 1 and Supplementary Figure 2), which is probably due to the divergence of ancient evolution between rice and Rauvolfia serpentine. By using the MEME motif search tool, we identified that: (1)  showed that OsSTRL19, OsSTRL20, and OsSTRL21 were expressed in all the tissues investigated. OsSTRL5 and OsSTRL6 showed a similar expression pattern of having a specific but weak transcription signal in the young panicles. The expression values of OsSTRL8 and OsSTRL11 were extremely low in all tissues, although they were expressed in the anthers. OsSTRL14 and OsSTRL15 were expressed in young panicles and in other vegetative organs (leaf, stem, and root). Interestingly, only OsSTRL2 showed a markedly specific transcript signal in both  1  OsSTRL1  LOC_Os01g50330 28892212-28891193  1  Clade II/group I   2  OsSTRL2  LOC_Os03g15710 8666146-8664064  3  Clade III/group I   3  OsSTRL3  LOC_Os03g53950 30929823-30927223  3  Singleton/group I   4  OsSTRL4  LOC_Os06g35950 20983248-20982674  6  Group II   5  OsSTRL5  LOC_Os06g41820 25081930-25080893  6  Clade I/group I   6  OsSTRL6  LOC_Os06g41830 25085537-25086526  6  Clade I/group I   7  OsSTRL7 LOC_Os06g41850  pre-emergency inflorescence, young panicles, and anthers. For further validation, we examined OsSTRL2 expression patterns by using semi-RT-PCR (Fig. 1B) and quantitative real time PCR (q-PCR) (Fig. 1C) in a range of rice organs, including vegetative tissues, namely, roots, stems, and leaves, and reproductive tissues, such as pistils and anthers, at different stages of development. The results showed an exclusive and obvious expression level of OsSTRL2 in the anther with glume lengths of 3.0 mm to 6.0 mm, indicating that OsSTRL2 might have a function in anther development.
Using the full-length amino acids of OsSTRL2 as a query, we performed a BLASTP search according to the Phytozome database (www.phytozome.net) to explore the potential function of OsSTRL2. Five putative homologs of OsSTRL2 were obtained in Vitis vinifera (Vv), Zea mays (Zm), Arabidopsis thaliana (At), Triticum aestivum (Ta), and Brassica napus (Bn). The homolog amino acid sequences, including STR1 and OsSTRL2, were subsequently aligned with Clustalw2 31 , and a phylogenetic tree was generated from the MEGA5 32 . The phylogenetic tree showed that OsSTRL2 and its related proteins, which were identified by Phytozome BLASTP, were grouped into the same clade, whereas only STR1 belonged to the other branch ( Fig. 2A). OsSTRL2 shared ~82.2%, ~79.9%, ~58.8%, ~60.5%, and ~60.5% identities with the OsSTRL2-related protein sequences in Ta, Zm, Vv, Bn, and Arabidopsis, respectively (Fig. 2). However, only 24.3% identity was found between OsSTRL2 and STR1 proteins (Fig. 2). The previous 3D-structural and functional analyses suggested that the six-bladed β-propeller and three α-helices formed the basic structure of the STR1 protein; some important residues were located in the STR1 substrate-binding region, in which the Glu-309 was experimentally proven to be the key catalytic residue 4 . Although the motifs for β-propeller folds and the residues forming a disulfide bridge that pulls two α-helices together were conserved 4-7 , several variations were found within the substrate-binding region in both OsSTRL2 and its homologues of the same branch, including the key site of Glu-309 (Fig. 2B). All OsSTRL members lack the residue Glu-309 which acts as the key catalytic residue of STR1 protein 4 (Supplementary Figure 1). Moreover, the critical residues that were not found in the OsSTRL2 protein were also absent in the two Arabidopsis strictosidine synthase-like (ATSsl) proteins, ATSsl7 and ATSsl14 (Supplementary Figure 4), which do not exhibit STR enzymatic activity 33 . These results suggested that OsSTRL2, which was specifically expressed in the developing anther, might be an atypical strictosidine synthase.

Knockout of OsSTRL2 in rice causes male sterility. The loss-of-function mutations in At Lap3 and Zm
Ms45, two putative orthologues of OsSTRL2 ( Fig. 2A), resulted in male sterility and defects in the pollen wall formation 34,35 , suggesting that OsSTRL2 might share conserved protein functions with LAP3 and MS45. To investigate the role of OsSTRL2 in rice development, we obtained 29 independent OsSTRL2 knockout plants by using the CRISPR/Cas9-mediated genome-editing tool. The target site in the leaves of T 0 transgenic lines were sequenced to determine the mutation of the target sequence (  Table 1). Among these mutants, four different mutation types were identified in the target site (  or biallelic mutations had short whitish anthers with very few pollen grains, which could not be stained by the I 2 /KI solution (Figs. 3E-G). By contrast, all heterozygous T 0 plants exhibited normal fertility as observed in the WT (Supplementary Table 1). Similar results were observed in several F2 populations generated by the crosses of these mutants and the WT, thereby further confirming the association of phenotype and genotype. The findings are as follows: (1) the binary construct for targeting the OsSTRL2 site had a high targeted editing efficiency, (2) the phenotype of male sterility in the knockout plants was caused by the mutations revealed in the gene OsSTRL2, and (3) OsSTRL2 played an essential role in controlling male fertility.

Defects of anther wall and pollen exine development in
OsSTRL2 knockout lines. We performed transverse section analysis for the anthers of the WT and the mutant to investigate the defects of pollen development in the knockout mutant of OsSTRL2. The previous study classified the rice pollen and anther development into 14 stages from the formation of stamen primordium to the release of mature pollen during anther dehiscence 36,37 . At stage 7, meiosis was initiated by the WT MMC within the locule that was surrounded by the four-layered anther wall of epidermis, endothecium, middle layer, and tapetum from the surface to the interior (Fig. 4A). MMC subsequently formed dyads and tetrads after two steps of meiosis (Figs. 4B and C). Towards the end of stage 8b, the tapetal cells initiated linearization and centralization, and tapetum had close cell arrangement 36,37 (Fig. 4C).
No detectable defects were observed between the WT and the mutant anthers until the anther stage 9 (Figs. 4A-C and E-G). However, the mutant anther showed clear morphological differences after stage 8b. At stage 9, the mutant microspores had a slightly wrinkled shape, although these were released normally as with WT (Figs. 4D and H). In stage 10, the WT microspores were vacuolated and became round-shaped. The tapetal cells started to deteriorate (Fig. 4I), and asymmetric mitosis occurred in vacuolated microspores, which formed generative and vegetative cells at the stage 11 (Fig. 4J). The tapetum of WT gradually thinned from stages 10 to 11( Fig. 4I-J). By contrast, the wrinkled microspores in the mutant had more observable shrinkage deformity from stages 10 to 11 (Figs. 4M,N). The delayed degradation of the tapetum was also observed in the mutant at the  (Figs. 4M,N). At stage 12, the microspores of the WT were round, enlarged, enriched in starch, and developed into mature pollens that filled the anther locule (Fig. 4K). However, the microspores of the mutant were extremely wrinkled, and the inner surface of mutant anther wall still had an incompletely degraded tapetum (Fig. 4O). At stage 13, the defected microspores of the mutant wrinkled into a distorted and vermiform shape, which thereby emptied the locules of the mutant anthers, whereas the WT anthers were filled with mature pollen and started dehiscence (Figs. 4L and P).
Scanning electron microscopy (SEM) was used during stage 12 anther samples of the WT and the mutant to obtain detailed information on the defects of the anther wall and pollen exine in the mutant. In agreement with the phenotypic observation results (Fig. 3E), the anthers of the mutant (Fig. 5B) were also shorter and smaller than those in the WT (Fig. 5A). In contrast to those with the WT (Figs. 5C and E), the wax crystals of the mutant anthers exo-surface were crowded and less regularly arranged, leading to a compact cell arrangement of anther epidermis (Figs. 5D and F). Orbicules are critical for exporting materials from tapetum to microspores 38 . However, compared with the smooth and regularly arranged orbicules emerging in the WT (Fig. 5G), slightly rough and congested orbicules were formed by the mutant inner locule side of tapetum surface (Fig. 5H). Consistent with the transverse section results (Fig. 4P), the mutant produced less, deflated, and adherent pollen grains (Fig. 5J), whereas the pollen grains in the WT were abundant and round-shaped (Fig. 5I). The exine, mainly composed of sporopollenin, is the most important outer layer of the pollen wall 39,40 . We scanned the surfaces of pollen grains in the WT and the mutant. The pollen grains in the WT showed elaborate exine patterning and roof-like tectum structure on their surface (Figs. 5K and M). By contrast, the exine surface was discontinuous and plush-like, and the roof-like tectum structure was generally lost in the mutant (Figs. 5L and N). These results suggested that OsSTRL2 was involved and played an important role in the network of regulating anther wall formation, tapetum degeneration and pollen exine development in rice.
OsSTRL2 was specifically expressed in tapetum and microspores. To directly determine the spatial expression pattern of OsSTRL2, we applied promoter-GUS reporter system to detect its transcriptional activity in the tissues examined with semi-RT-PCR and q-PCR analyses. The transgenic lines containing a GUS-reporter gene driven by the OsSTRL2 promoter further confirmed the specific expression of OsSTRL2 in the developing anther (Figs. 6A-F). Particularly, the section of the GUS-staining anther showed that OsSTRL2 was expressed in the tapetum and microspores (Fig. 6H). RNA in situ hybridization was performed using the WT anther sections to precisely investigate the spatial and temporal expression patterns of OsSTRL2. The results revealed that OsSTRL2 expression was strong and specifically occurred in the tapetum at stage 8 (Fig. 6J). All these results support the functional role of OsSTRL2 in regulating pollen development.

OsSTRL2 protein is primarily localized to the ER. To determine the subcellular localization of
OsSTRL2, we generated a yellow fluorescent protein (YFP) fused to the C-terminal of OsSTRL2, under the control of the cauliflower mosaic virus double 35 S promoter (2 × 35S::OsSTRL2-YFP). Transiently expressed results showed that the YFP signal was localized at the endoplasmic reticulum (ER) that surrounds the nuclei (ER-ring) in the tobacco leaf epidermal cells (Figs. 7B and C) and 2 × 35S:: YFP as the control (Fig. 7A). Moreover, YFP signals were also detected on some tubular and cisternal patterns similar to the ER-like structure (Figs. 7B and C). The co-infiltration of Agrobacterium containing the ER-marker and 2 × 35S::OsSTRL2-YFP to the epidermal cell layers of tobacco leaves was conducted to confirm the complex localization of OsSTRL2. Confocal laser-scanning microscopy indicated that the YFP signals (Fig. 7D) were well merged with the ER red fluorescent protein (RFP) signals (Figs. 7E and F). Thus, OsSTRL2 proteins were primarily localized in the ER-ring and in the tubular and cisternal ER structures. These results also suggested that the ER might be an important site for the regulation of pollen exine formation.

Discussion
Strictosidine is the common precursor of all indole alkaloids and is synthesized from tryptamine and secologanin under STR catalysis, which is the key step of TIA biosynthesis 1, 3 . The plant STR is expressed in active meristematic tissues, such as young leaves, flower buds, shoot tips, and petals during plant development, and has a close connection with cell division 41 .
The genome-wide analysis showed that rice contains at least 21 OsSTRL genes that were divided into two groups (Table 1 and Supplementary Figure 2). The large family members of OsSTRL indicated their essential role in rice growth and metabolism. The expression patterns analysis of OsSTRL genes from public data and semi RT-PCR ( Fig. 1 and Supplementary Figure 4) showed that five OsSTRL members (OsSTRL3, OsSTRL12, OsSTRL19, OsSTRL20, and OsSTRL21) were constitutively expressed in all tissues, whereas nine members (OsSTRL1, OsSTRL4, OsSTRL7, OsSTRL9, OsSTRL10, OsSTRL13, OsSTRL16, OsSTRL17, and OsSTRL18) displayed extremely weak transcription signals in any of the tissues investigated, which suggests that these genes might be expressed at specific developmental stages or under special conditions. However, only OsSTRL2 had strong and specific transcription signals in pre-emergency inflorescence, young panicles, and anther (Fig. 1A). Similar to male sterility-related genes that were recently identified, OsSTRL2 was preferentially expressed in the developing anther before mature pollen formation as suggested by the additional investigation via semi-RT-PCR and q-PCR (Figs. 1B and C).
Protein sequence alignment analysis reveals that the OsSTRL2 possesses high identities of 79.9% and 60.5% of MS45 (maize) and LAP3 (Arabidopsis), respectively (Fig. 2). Both MS45 and LAP3, also annotated in the STR family, are specifically expressed in their anthers, and the mutants of these two genes are male sterile 34,35 . In Tapetum, the innermost cell layer of anther wall, produces sporopollenin precursors that are transported to the developing microspores for pollen exine formation [42][43][44] . Ubisch body, also named orbicule, is located on the inner surface of the tapetal cells and is only a few microns (μm) in size 45 . Ubisch body is one of the most essential by-products in pollen wall sporopollenin synthesis 46 . Within the OsSTRL2 knockout mutant anther, tapetum degeneration was delayed (Figs. 4M and N), and the ubisch bodies in the mutant inner locule side of tapetal surface had irregular shape and chaotic arrangement compared with those in the WT (Fig. 5H). Moreover, epidermal waxes can primarily protect the land plants from water loss and can also be involved in plant defenses against pathogens 47,48 . Compared with the WT anther, the crowded permutation and distorted appearance of wax crystals on the mutant outer epidermis, might be the main reason for the smaller anther exo-surface cell size in the mutant anther (Figs. 5C-F). These results indicated that OsSTRL2 has a key function in the biological pathway of tapetum PCD and ubisch body formation. OsSTRL2 loss-function might affect the expansion and morphology of the anther epidermal cells by producing abnormal wax crystals.
Pollen wall is a surrounded lipidic structure of male gametophyte that plays an essential role in protecting pollen from various environmental stress and bacterial attack. Pollen wall consists of two layers: the outer exine, which contains tectum and baculum; and the inner intine 39,40,49,50 . In Arabidopsis, PKSA and PKSB, two plant type III polyketide synthases (PKSs), mediate the biochemical reactions to induce the synthesis of pollen fatty acids and phenolics found in exine 51,52 . As an Acyl-CoA synthetase catalyzing the fatty acyl-CoA ester biosynthesis reaction in Arabidopsis tapetal cells, ACOS5 plays a central role in generating sporopollenin monomers that can be exported to microspores and regulate the formation of pollen exine 53,54 . LAP3 is a homolog of OsSTRL2 in Arabidopsis (Fig. 2). Lap3 mutant plants produce less adherent pollen grains because of their thinner pollen exine walls 34 . Similar to that in lap3, the OsSTRL2 knockout mutant pollen grains displayed structurally weakened exine that lacks tectum (Fig. 5N). This defect might cause pollen grains to rupture easily and cause adhesion (Figs. 3G, 4M-P and 5J). By contrast, the pollen grains in WT were round and arranged by normal roof-like tectum on their outer exine surface (Fig. 5M). These results suggested that OsSTRL2 might use similar regulatory pathways for pollen exine formation in Arabidopsis.
The mutation in Ms45 gene, a maize homolog of OsSTRL2 (Fig. 2), causes the absence of pollen. Immunological analyses showed that MS45 was localized to the tapetal cells during microsporogenesis 35 . Our results indicated that OsSTRL2 was specifically expressed in tapetum and microspores via GUS reporter analysis and RNA in situ hybridization (Fig. 6), suggesting that OsSTRL2 and MS45 might share this tapetal localization and play the conserved roles during exine biosynthesis. PKSA, PKSB, and TKPR1 are required for sporopollenin biosynthesis in Arabidopsis 51, 52, 55 . These enzymes are immunolocalized to the ER of anther tapetal cells 56 . Similarly, our results also showed that OsSTRL2 was mainly localized in the ER (Fig. 7), whereas the STR1 orthologue in periwinkle was simultaneously targeted to the chloroplast, vacuole and ER in transgenic tobacco 57 . These results suggested that the ER of tapetum might be an important site for the proteins regulating pollen wall formation, and therefore OsSTRL2 might participate in this essential network by influencing the tapetum degeneration, anther wax crystal development and pollen exine formation (Fig. 8).
The structural characterizations of STR1 deciphered the 3D structure of the active site of strictosidine synthase and the details of its reaction mechanism 3, 4, 7 . The structure of STR1 contains six-bladed β-propellers and three α-helices. The substrate-binding pocket of STR1, in which several vital residues (Fig. 2B) are involved, links the active center to the surface of the substrate molecule. In addition, the structural analysis and site-directed mutagenesis experiments found that Glu-309 (Fig. 2B) in the residues of STR1 binding pocket is a key catalytic residue 4,7 . Among the typical STR family, these motifs and residues are important and conserved because they provide the basic structure of the protein 7 . However, OsSTRL2 lacks Glu-309, which is the key catalytic residue of typical strictosidine synthase STR1. OsSTRL2 also exhibits differences in some residues in the substrate-binding pocket (Fig. 2B), although it contains the motifs for β-propeller folds and the residues Cys-89 and Cys-101, which form a disulfide bridge and pull two out of three α-helices together (Fig. 2B). This finding suggested the possibility that OsSTRL2 might have no or have low catalytic activity as presented in the typical strictosidine synthase. Similar to OsSTRL2, AtSsl7 and AtSsl14 are Arabidopsis strictosidine synthase-like proteins that lack the Glu-309 and some other residues in the substrate-binding region (Supplementary Figure 4) and do not have strictosidine synthase activity 33 . This result might offer experimental evidence to our hypothesis on the catalytic activity of OsSTRL2.
LAP3 is the homologous gene of OsSTRL2 in Arabidopsis (Fig. 2) and is one of the ATSsl genes 4 . LAP3 has a relatively low similarity to the protein sequence of STR1 34 . Our results also revealed a comparatively low amino acid sequence identity between OsSTRL2 and STR1 (Fig. 2). These findings indicated that both LAP3 and OsSTRL2 have evolutionary divergence with STR1. In addition, Kibble et al. found that all members of the ATSsl family lack the Glu-309 and suggested that this family might not have strictosidine synthase enzyme activity 4 . The absence of this critical catalytic residue also occurs in all OsSTRL family members of rice (Supplementary Figure 1). These differences might bring variations in the substrate specificity and/or enzymatic activity for OsSTRL family members, as well as the possibilities that those different members might participate in other biochemical pathways in plants 11 . Although OsSTRL2 lacks some the key residues of STR, we propose that OsSTRL2 is an atypical strictosidine synthase because it processes the basic framework of a STR protein (e.g., β-propeller folds and α-helices).
In summary, this study identified 21 rice OsSTRL genes through genome-wide analysis. The 21 OsSTRL members were clustered into two major groups and four clades in group I. The expression pattern analysis showed that only OsSTRL2 was specifically expressed in the developing anther. OsSTRL2 loss-function in rice caused the detention of tapetum degeneration, abnormal wax crystals, and orbicules of anther and the absence of pollen exine tectum, which led to the knockout mutant male sterility. Based on the accumulation of OsSTRL2 in tapetal cells and microspores at the transcript level and the ER localization of OsSTRL2-YFP protein, OsSTRL2 might participate in the regulation network of pollen wall formation that occurs in the ER of tapetum. By combining the results of protein sequence alignment, we consider OsSTRL2 as a rice atypical strictosidine synthase that play key roles in anther development and pollen wall formation. Our work also provides a new perception into mechanisms of anther and pollen developmental processes and enhances our understanding of the regulation of pollen exine formation.

Materials and Methods
Plant materials and growth conditions. All rice (Oryza sativa) plants used in this study were naturally grown and maintained in the experimental field of the Rice Research Institute, Sichuan Agricultural University, Wenjiang, China.
Genome-wide sequence analysis of rice STR-like genes and phylogenetic analysis. Using the full-length STR1 protein sequence in Rauvolfia Serpentine as a query, a BLASTP search of the japonica-type rice genome (RGAP; http://rice.plantbiology.msu.edu/) 28 was performed to forecast the 21 OsSTRL candidates. The amino acid sequences of these candidates were subsequently used for the SMART search 29,58 and were aligned by using Clustal W software (www.ebi.ac.uk/Clustalw) to further confirm the accuracy. The OsSTRL2-related protein sequences in different species were identified by Phytozome Blast (https://phytozome.jgi.doe.gov) with default parameters using the full-length amino acid of OsSTRL2 as a query, and the sequences aligned with Clustalw2 31 . In addition to the 1,000 bootstrap replications, the MEGA5 program used the NJ method with default parameters to generate all the phylogenetic trees 32 .
CRISPR/Cas9-mediated mutation and phenotype association assay. To verify the OsSTRL2 function, we generated a gRNA construction, wherein the gRNA and the plant-optimized Cas9 were driven by the rice U3 and maize UBI promoter, respectively 59 . Plasmid was introduced into the WT (Nipponbare). The DNA isolated from the leaves of transgenic plants was subjected to PCR and sequencing analysis with the primer set OsSTRL2-gRNA-seq (see Supplementary Table S3) to evaluate whether the mutation occurred. To verify the association between the candidate mutation site in OsSTRL2 and the male sterile phenotype, we observed the target site sequence of all the CRISPR/Cas9 transgenic plants via direct or cloned sequencing of the PCR products, which were amplified using the primer set OsSTRL2-gRNA-seq (statistical results are shown in Supplementary Table 1). The co-segregation results were further confirmed in several F2 populations that were generated by backcrossing these mutants with the WT.
Phenotypic characterization. Transverse sections of the anther development analysis were performed as described previously 19 . For the SEM analysis of anther and pollen, the samples at stage 12 were prepared as described by Qin et al. 25 and then examined with a JSM-7500F scanning electron microscope.
Expression analysis. Digital expression pattern analysis of 21 OsSTRL members was performed by using the public RNAseq data from NGSTD (http://rice.plantbiology.msu.edu/expression.shtml) to draw a heat map with the OmicShare tools, a free online platform for data analysis (http://www.omicshare.com/tools/). Total RNAs from a variety of rice tissues were extracted using the TriPure isolation reagent (Roche, Indianapolis, USA). cDNAs were reverse-transcribed using the Transcriptor First-Strand cDNA Synthesis Kit (Roche, Indianapolis, USA). Semi-RT-PCR was performed with the program involving pre-denaturation at 95 °C for 2 min, 29 or 30 cycles of the reaction at 95 °C for 30 s, 57 °C for 30 s and 72 °C for 30 s. Another step of at 72 °C for 7 min was conducted for the final extension, and the PCR amplification results were run on 1% agarose gel. The gel image was obtained by using Molecular Image Gel Doc XR + image analysis system (Bio-Rad, Hercules, CA). qPCR experiments conducted by using a Bio-Rad CFX96 real-time PCR System (California, USA) as described by Li et al. 60 , the rice actin gene was used as the internal control, and the fold change for gene expression was calculated as described previously 61 . All primers used in this study are listed in Supplementary Table 3.
Promoter-GUS reporter assays and RNA in situ hybridization. For the transcript level tissue localization of OsSTRL2, a construct of OsSTRL2pro::GUS was generated through pCAMBIA1300 wherein the native OsSTRL2 promoter drove the GUS gene. The OsSTRL2pro::GUS construct was introduced into the Agrobacterium tumefaciens strain EHA105 and then transferred into the WT. Histochemical GUS staining was performed as described previously 62 . A 406 bp fragment of OsSTRL2 cDNA was amplified using specific primers (Supplementary Table 3) and cloned into pBluescript-SK vector (Stratagene) to generate the OsSTRL2-specific antisense probe. The fragment was transcribed with T7 RNA polymerase. In situ hybridizations including tissue embedding, hybridization and signal detection were conducted as described by Chen et al. 63 .
Subcellular localization of OsSTRL2. The full-length cDNA of OsSTRL2 was cloned into pA7-YFP vector to generate a 2 × 35S::OsSTRL2-YFP cassette which fused the C-terminal of OsSTRL2 with YFP under the control of the cauliflower mosaic virus double 35 S promoter. The entire 2 × 35 S::OsSTRL2-YFP cassette was then inserted into the pCAMBIA1300. The RFP with the C-terminal extension-KDEL served as the ER marker under the control of the constitutive cauliflower mosaic virus double 35 Spromoter 64 . These plasmids were individually expressed or co-expressed in tobacco leaf epidermis cells by agrobacterium-mediated infiltration. YFP and RFP signals were visualized with a confocal scanning microscope (Nikon A1, Kanagawa, Japan) 72 h after infiltration.