SCR-22 of pollen-dominant S haplotype class is recessive to SCR-44 of pollen-recessive S haplotype class in Brassica rapa

SCR/SP11 encodes the male determinant of recognition specificity of self-incompatibility (SI) in Brassica species and is sporophytically expressed in the anther tapetum. Based on dominance relationships in pollen and nucleotide sequence similarity, the S haplotypes in Brassica have been classified as class I or class II, with class-I S haplotypes being dominant over class-II S haplotypes. Here, we revealed that S-22 in B. rapa belonging to class I is recessive to class-II S-44 and class-I S-36 in pollen, whereas it is dominant over S-60, S-40, and S-29 based on pollination tests. SCR/SP11 of S-22 (SCR-22) was sequenced, revealing that the deduced amino-acid sequence of SCR-22 has the longest C-terminal domain among the SCR/SP11 sequences. The expression of SCR-22 was found to be suppressed in S-22/S-44 and S-22/S-36 heterozygotes. Normal transcription of SCR-44 was considered to be due to the transcription suppression of Smi sRNA of the S-22 haplotype and a very low methylation state of the SCR-44 promoter region in the tapetum of S-22/S-44 heterozygotes. In SCR-22, only the cytosine residue located at the –37 bp position of the promoter region was hypermethylated in the tapetum of S-22/S-44 heterozygotes, and few methylated cytosines were detected in the promoter and coding regions of SCR-22 in S-22/S-36 heterozygotes. SCR-22 was also expressed in microspores in S-22 homozygotes but not in S-22/S-44 and S-22/S-36 heterozygotes. These results suggest that a mechanism different from class-II SCR/SP11 suppression may operate for the suppression of recessive class-I SCR-22 in S heterozygotes. Studies in cabbage plants shed new light on self-incompatibility mechanisms to avoid self-fertilization. In Brassica, self-incompatibility is controlled by the expression of dominant and recessive genetic variants in the S-locus region. These variants encode proteins that mediate the rejection of self-pollen, but little is known about how the dominance relationship is established between them. Takeshi Nishio at Tohoku University in Japan and colleagues have examined the sequence, expression pattern and dominance relationships of the gene encoding the pollen-coat protein SCR-22. Despite sharing features with dominant SCR variants, SCR-22 can act in a recessive manner. Interestingly, unlike other recessive SCR variants, the suppression of SCR-22 expression does not depend on the addition of methyl groups that prevent transcription factor binding. This finding suggests the dominance hierarchy is governed by different mechanisms.


Introduction
Self-incompatibility (SI) is a genetic mechanism exploited by many angiosperm species to prevent inbreeding and to promote outcrossing. In most species, SI is controlled by a single S locus with a large number of haplotypes. The SI response occurs when the S haplotype of the pollen is the same as that of the pistil. In Brassica, three genes located at the S locus have been characterized, namely, S-receptor kinase (SRK), a female determinant of recognition specificity; 1,2 S-locus cysteine-rich protein/S-locus protein 11 (SCR/SP11), a male determinant; 3,4 and S-locus glycoprotein (SLG), which is highly similar to the extracellular domain (S-domain) of SRK 5 . SCR/SP11 is a small cysteine-rich protein with ca. 50 amino-acid residues with 8 conserved cysteine residues. The sequence of mature SCR/SP11 protein is highly polymorphic, with less than 50% amino-acid similarity among different S haplotypes within the same species 3,6-9 . In most SCR/SP11 variants, only a few amino acids are conserved, such as the eight cysteines, a glycine between the first and second cysteines, and an aromatic aminoacid residue between the third and fourth cysteines 3,6,7,10 . SCR/SP11 proteins are mainly produced in the anther tapetum and are then transferred to the surface of mature pollen 3,10-12 . Therefore, the SI phenotype in pollen is consistent with the dominant S haplotypes carried by S heterozygous plants.
Based on dominance relationships relative to the other alleles in S-heterozygous plants and the nucleotide sequence similarity of S-locus genes, the S haplotypes in Brassica are classified into two groups: the pollendominant S haplotypes termed class I and the pollenrecessive S haplotypes termed class II 13 . The class-I S haplotypes are always dominant over the class-II S haplotypes in pollen. In Brassica rapa, SCR/SP11 sequences of four class-II members, namely, S-44, S-60, S-40, and S-29, have been identified, and a linear dominance relationship has been demonstrated among them 14,15 . The expression patterns of class-I and class-II SCR/SP11 alleles are slightly different. Class-II SCR/SP11 alleles are expressed only in the anther tapetum, whereas class-I SCR/SP11 alleles are expressed not only in the tapetum but also in microspores 3,10,16 . Phylogenetic analysis has revealed that class-II SCR/SP11 alleles form a distinct group separate from class-I SCR/SP11 alleles 15 . The promoter sequences of SCR/SP11 alleles show little similarity between the two classes, which results in an expression pattern difference between class-I and class-II SCR/SP11 alleles.
In heterozygotes with class-I and class-II S haplotypes, the class-II SCR/SP11 is not expressed, indicating that the dominance relationships are regulated at the messenger RNA level of SCR/SP11 alleles 15 . It was found that class-I SCR/SP11 alleles having promoter defects, which are not transcribed, also caused suppression of recessive class-II SCR/SP11 alleles 17 . Subsequent studies demonstrated that the expression suppression of recessive class-II SCR/SP11 alleles results from their tissue-specific methylation of promoter sequences in the tapetum 18 . Further analysis showed that a sequence with high similarity to the target methylated region lies in a region flanked by dominant SLG alleles named SP11-methylation-inducer (Smi). An Smi sequence was used as a template to form a 24-nucleotide small noncoding RNA (sRNA), which induced the methylation of the promoter of a recessive SCR/SP11 allele and repressed its transcription 19 . Recently, Smi2 has been identified in a class-II S haplotype sequence and has been shown to control the linear dominance hierarchy of the four class-II SCR alleles 20 .
S-22 in B. rapa has been reported to be recessive to S-24, S-26, and S-43, belonging to class-I S haplotypes, in pollen in the same way class-II S-60 is recessive to S-24, S-26, S-28, and S-43 21 . In the stigma, S-22 has been revealed to be recessive to S-28 in class-I S haplotypes. The nucleotide sequences of SLG and SRK of S-22 have been determined and deposited in the DDBJ (AB054060 and AB054061, respectively). Comparison of the nucleotide sequences and deduced amino-acid sequences of these alleles with those of other SLG and SRK alleles has revealed that S-22 belongs to the class-I S haplotypes 8 . Nucleotide sequences of SCR/SP11 of S-22 have not been reported. Since S-22 is ranked the lowest in the dominance hierarchy of S haplotypes among class-I S haplotypes, this haplotype may have some unique characteristics. In the present study, we found that S-22 was recessive to S-44 in class-II haplotypes in pollen. We identified SCR/SP11 of S-22 (SCR-22 hereafter), finding that it has a unique feature in deduced amino-acid sequences. The expression of SCR-22 was suppressed in the S-22/S-44 heterozygote, but the cytosine methylation pattern in the SCR-22 promoter was different from that of recessive class-II SCR alleles. These results may help us to better understand the mechanism controlling dominance relationships among SCR/SP11 alleles.

Results
Dominance relationships between S-22 and other S haplotypes of B. rapa in pollen We analyzed the dominance relationships in pollen between S-8, S-22, and class-II S haplotypes of B. rapa by pollination tests. The class-I allele S-8 was used as a  Sequence analysis of SCR-22 of B. rapa We determined the nucleotide sequence of the coding region of SCR-22 and its promoter region sequence in two steps. Since SCR-22 was not amplified by reverse transcription-polymerase chain reaction (RT-PCR) using the primers reported by Watanabe et al. 6 and Sato et al. 8 , we amplified a partial sequence of SCR-22 of B. rapa using many combinations of primers including newly designed primers (Supplementary Table S1). Second, we amplified the flanking sequence of the identified region of SCR-22 by inverse PCR to determine the nucleotide sequence of the whole coding region of SCR-22 and its promoter region. Our results showed that the coding region of SCR-22 is 627 bp in length and contains a 306 bp intron (Supplementary Figure S1).
It was found that three amino acids, i.e., the seventh, tenth, and twelfth amino acids, in the putative signal peptide of SCR-22 are different from those of other class-I SCR sequences (Fig. 1). As with other SCR/SP11 proteins, the eight conserved cysteine residues are present in SCR-22, and a glycine residue between C1 and C2 is also conserved in SCR-22. However, the length of the SCR-22 protein is different from that of other SCR/SP11 proteins. Most SCR/SP11 proteins contain approximately 50 amino acids, whereas the SCR-22 protein contains approximately 70 amino acids. SCR-22 has a longer C-terminal domain (Fig. 1). Linkage analysis showed that this gene was linked to SRK-22 in B. rapa (Supplementary Figure S2). These results confirm that the gene we identified is SCR-22 of B. rapa.

Tapetum isolation
It has been reported that class-II SCR/SP11 alleles are specifically transcribed in the tapetum 15 and that DNA methylation of the promoter of recessive SCR/SP11 alleles in heterozygotes also occurs in the tapetum 18 . In previous studies, DNA was extracted from the nuclei of tapetum cells, but the method was not described in detail 18 . We therefore developed a method for tapetum isolation. The anther contains endothecium, tapetum, and microspores ( Fig. 2a). We cut away both ends of the anthers and released the microspores by vortexing. All microspores were released (Fig. 2b). The anthers were then treated with cellulase RS and pectolyase Y23 for only 7 min to release the tapetum cells (Fig. 2c). We then used this method to isolate the tapetum cells from the anthers of S-60 homozygotes and analyzed the expression of SCR-60 in isolated endothecium, microspores, and tapetum (Fig. 2d). The expression of SCR-60 was detected only in the isolated tapetum cells, not in isolated endothecium or microspores, confirming that the fraction we obtained was the tapetum and that few tapetum cells were present in the isolated endothecium fraction or microspore fraction. The tapetum cells from the anthers of S-22/S-60 heterozygotes were then isolated to assess the methylation rate of the recessive SCR-60 5' region. Methylated cytosine residues at CpG, CpNpG, and CpNpN sites were widespread, and the methylation frequencies of the two cytosine residues in the region of SCR-60 homologous to the Smi of the S locus 19 were 46.9% and 71.9%, respectively (Fig. 2e). The highest methylation frequency of cytosine residues in the 5' region of class-II SCR/SP11 alleles has been reported to be approximately 80% 18,19 . Considering that the DNA methylation of SCR/SP11 alleles is detected only in the tapetum and that the highest methylation frequency detected in the present study was close to the highest reported methylation frequency, it can be inferred that the purity of the isolated tapetum fraction prepared using our method is comparable to that obtained by the method of Shiba et al. 18 and that the isolated tapetum fraction of our study can be used for the following analyses.

Gene expression of SCR/SP11 alleles in S heterozygotes
The relative expression level of each SCR/SP11 allele was investigated in the S heterozygotes. First, the expression level of SCR  (Fig. 3b). SCR-8, also belonging to class I, was expressed in the tapetum cells of all heterozygotes we analyzed (Fig. 3c). Second, the relative expression levels of four class-II SCR/SP11 alleles were investigated. SCR-44 was not expressed in the tapetum cells of S-8/S-44 heterozygotes but was expressed in those of S-22/S-44 heterozygotes (Fig. 3d). SCR-60, SCR-40, and SCR-29 were not expressed in the tapetum cells of any of the S heterozygotes we analyzed (Fig. 3e, f, g). SCR-36 was expressed in the tapetum cells of S-22/S-36 heterozygotes (Fig. 3h). The observed relative expression levels suggested that SCR-22 is dominant to SCR-60, SCR-40, and SCR-29 and recessive to SCR-44 and SCR-36 and that SCR-8 is dominant to all the class-II SCR/SP11 alleles that we analyzed. These results were consistent with the results of the pollination tests.  The methylation state of recessive SCR alleles in heterozygotes It has been reported that suppression of the expression of recessive class-II SCR/SP11 alleles results from methylation of the promoter region of recessive SCR/SP11 alleles induced by an sRNA of the class-I or class-II S haplotype in the tapetum 19,20 . The methylation state of SCR-44 was therefore measured in the present study. Widespread methylated cytosine residues were found in the SCR-44 promoter region in the tapetum of S-8/S-44 heterozygotes (Fig. 4a). All three types of cytosine methylation, i.e., CpG, CpNpG, and CpNpN, were detected in this region. The methylation frequencies of two cytosine residues in the region homologous to Smi in the SCR-44 promoter of S-8/S-44 heterozygotes were 32.1% and 39.3%, respectively. In the tapetum of S-22/S-44 heterozygotes and S-44 homozygotes, where SCR-44 is transcribed, few methylated cytosine residues were detected in the promoter region of SCR-44 (Fig. 4a). In addition, methylated cytosine residues in the region homologous to Smi in recessive SCR-60, SCR-40, or SCR-29 were also observed, and the percentages of methylated cytosine were from 21.7% to 73.7% in the heterozygotes, which are higher than the 2.8% to 11.8% observed in the homozygotes (Supplementary Figure S3). These results indicate that the suppression of class-II SCR/SP11 expression is related to DNA methylation.
The methylation states of SCR-22 in S-22 homozygotes and S-22/S-44 and S-22/S-36 heterozygotes were also investigated. In the tapetum of S-22 homozygotes, the methylation rate of the SCR-22 promoter region was very low (Fig. 4b). In the tapetum of S-22/S-44 heterozygotes, where SCR-22 was not transcribed, the cytosine residue located at the -37 position of the promoter was found to be highly methylated, with a methylation rate of 74% (Fig. 4b). No further methylated cytosine was detected in the SCR-22 promoter region, coding region, and intronic region in the tapetum (Fig. 4b, c). At the same time, few methylated cytosines were detected in the SCR-22 promoter region, coding region, and intronic region in the microspores of S-22/S-44 heterozygotes (Fig. 4d), where SCR-22 was also not transcribed. Additionally, in S-22/S-36 heterozygotes, which carry two class-I S haplotypes, a methylated cytosine-rich region was detected at the promoter region (from -350 to -440) of SCR-22 in the tapetum. However, the methylation rates of these cytosine residues were low, with the highest methylation rate being 31.2%, and only two types of cytosine methylation, i.e., CpNpG and CpNpN, were detected in this region. Few methylated cytosine residues were detected in the coding and intronic regions of recessive SCR-22 (Fig. 4b-d). In the microspores of the S-22/S-36 heterozygotes, similar to those of S-22/S-44 heterozygotes, few methylated cytosines were detected in the SCR-22 promoter region, coding region, and intronic region (Fig. 4b-d).
Transcript analysis of Smi trans-acting sRNA It has been reported that the Smi trans-acting sRNA from the class-I S locus induces the methylation of the promoter of recessive class-II SCR/SP11 alleles 19 . The primer set SL-F1/SL-R1, designed by Tarutani et al. 19 , was used to amplify the sequence of the precursors of Smi-8 and Smi-22 from the S locus of S-8 and S-22 haplotypes. The results showed that the precursors can form an imperfect stem-loop structure (Fig. 5a), and the sequences of the Smi-8 and Smi-22 sRNAs are the same as that of the Smi-9 sRNA 19 .
The  (Fig. 5b). Furthermore, the expression of precursors of Smi-22 was not detected in S-22/S-44 heterozygotes (Fig. 5c). Mature sRNA of Smi was also detected in all plants having the S-8 haplotype that we analyzed, including S-8/S-44 heterozygotes (Fig. 5d). To confirm that the mature sRNA of Smi is formed from the class-I S locus, not from the class-II S locus, the expression level of mature sRNA of Smi was also analyzed in S-44, S-60, S-40, and S-29 homozygotes. The expression of mature sRNA was hardly detected in the S-29 homozygotes and was not detected in the S-44, S-60, and S-40 homozygotes (Fig. 5e), confirming that the mature sRNA detected in the heterozygotes was mainly from the class-I S haplotypes. These results suggest that SCR-44 expression in S-22/S-44 heterozygotes is due to the suppression of Smi expression.

Discussion
In the present study, we determined the nucleotide sequence of SCR-22 of B. rapa. SCR-22 was found to be 20 amino acids longer than other SCR/SP11 proteins. Such a long SCR/SP11 protein has not previously been reported. In the putative signal peptide of SCR-22, three amino acids, i.e., the seventh, tenth, and twelfth, were not conserved. These changes are not considered to contribute to the difference in the hydrophobicity level of the SCR-22 signal peptide from that of other class-I SCR/ SP11 proteins. The grand average of the hydropathicity value of the SCR-22 signal peptide was between those of SCR-8 and SCR-12, indicating that the function of the SCR-22 signal peptide was maintained.
SCR-44 of class-II SCR/SP11 alleles has been thought to be recessive to all class-I SCR/SP11 alleles. In the present study, pollination tests showed that class-I SCR-22 is recessive to SCR-44 but dominant to SCR-60, SCR-40, and SCR-29 (Table 1). The dominance relationships between SCR-22 and class-II SCR/SP11 alleles were further confirmed by gene expression analysis. The expression level of recessive SCR/SP11 alleles is greatly reduced in S heterozygotes 15 . Our results showed that the expression of SCR-22 is suppressed and that SCR-44 is normally expressed in the tapetum of S-22/S-44 heterozygotes (Fig. 2). Since the suppression of recessive class-II SCR/ SP11 transcription is considered to result from methylation of the promoter region induced by Smi, which can be observed in the tapetum 19,20 , we developed a method for tapetum isolation. Cytosine methylation was detected at a level comparable to that reported previously; 18,19 therefore, our tapetum isolation method was found to be usable for analyses of the methylation states of the SCR-22 promoter and the expression level of Smi.
Our investigation of dominance relationships showed that SCR-44 was recessive to class-I SCR-8 but dominant to class-I SCR-22. Methylation state analysis showed that the widespread methylated cytosine residues were present in the SCR-44 promoter in S-8/S-44 heterozygotes (Fig. 4). The methylation profile of the SCR-44 promoter in S-8/S-44 heterozygotes is similar to that of the SCR-60 promoter in S-52/S-60 heterozygotes 18,19 . The methylated cytosine residues of the SCR-44 promoter at CpG, CpNpG, and CpNpN sites suggest that Smi sRNA triggers monoallelic de novo methylation in the recessive SCR-44 promoter 19,22 . The methylation percentage of the SCR-44 promoter in S-8/S-44 was clearly lower than that of the SCR-60 promoter in S-52/S-60 heterozygotes and higher than the SCR-60 promoter in S-44/S-60 heterozygotes 18,19 . The methylation frequencies of two cytosine residues in the region homologous to Smi in the SCR-44 of S-8/S-44 heterozygotes were 32.1% and 39.3%, respectively. Methylated cytosine was also detected in the region homologous to Smi in the recessive SCR-60, SCR-40, or SCR-29 in the heterozygotes (Supplementary Figure S3). In the homozygotes with class-II SCR/SP11 alleles and S-22/S-44 heterozygotes, few methylated cytosines were observed in the promoter of class-II SCR/SP11 alleles. These results confirmed the suppression of recessive class-II SCR/SP11 alleles induced by DNA methylation in their promoter regions.
Because Smi has been revealed to play a key role in inducing the promoter DNA methylation of recessive class-II SCR/SP11 alleles, the precursor sequence of Smi from S-8 and S-22 was identified (Fig. 5a). The sequences of the mature Smi-8 and Smi-22 sRNAs were the same as that of the Smi-9 sRNA 19  homozygotes and was hardly detected in S-29 homozygotes. These results confirm that the Smi sRNA detected in the present study was mainly from S-8 or S-22. In addition, the precursors of Smi-22 were not detected in S-22/S-44 heterozygotes (Fig. 5c). Thus, the absence of Smi sRNA in S-22/S-44 heterozygotes is considered to be due to the transcription suppression of Smi-22 rather than a failure to cleave precursors of Smi-22 into mature sRNA. In addition, in S-22/S-44 heterozygotes, few  23 . Stress-responsive elements and tissue-specific regulatory elements have been found in the promoters of miRNA genes 24,25 . Smi is expressed in the tapetum specifically 19 , indicating that tissue-specific regulatory elements are present in the promoter region of the Smi-22 gene. Transcription of Smi-22 may be suppressed by some factor present in the S-locus sequence of S-44. The sequence of more than 10 kb of the S-44 haplotype has been determined and published 26 , but it is not available in the sequence database. Therefore, repeated sequencing analysis is required for identification of the factor responsible for the suppression of SCR-22 transcription.
Recently, Smi2 was identified to control the linear dominance hierarchy of the four class-II SCR alleles 20 . It is possible that SCR-22 expression is suppressed by Smi2 of the S-44 haplotype. However, no region similar to Smi2 was found in the promoter region of SCR-22, indicating that the suppression of SCR-22 expression in S-22/S-44 heterozygotes was not related to the Smi2 of SCR-44.
To examine whether the same suppression mechanism as that for class-II SCR/SP11 participates in the suppression of recessive class-I SCR-22 in S heterozygotes, the methylation state of recessive SCR-22 in the tapetum was analyzed. The cytosine residue located at the -37 position of the SCR-22 5' region was highly methylated in S-22/S-44 heterozygotes (Fig. 4). In recessive class-II SCR alleles, methylated cytosine residues are widespread in the promoter region, with all three types of cytosine methylation, i.e., CpG, CpNpG, and CpNpN, occurring in the region 18,19 . However, in recessive class-I SCR-22 in the tapetum of S-22/S-44 heterozygotes, methylated cytosine was restrictedly localized at the -37 cytosine, and only CpNpN methylation was observed. In S-22/S-36 heterozygotes, low levels of CpNpG and CpNpN methylation were detected in the region from -350 to -440 bp of the SCR-22 5' region. Promoter hypermethylation around cis-regulatory elements could affect transcription repression by interfering with the transcription machinery 27,28 . It has been reported that the region around -192 bp of SCR-9 contains the elements required for expression in the tapetum 12 . Alignment revealed that the 5' region between -1 and -200 bp of SCR-22 and SCR-9 was highly conserved (Supplementary Figure S4), with the two sequences sharing 82.7% identity. Thus, the region around -191 bp of SCR-22 is inferred to contain the elements required for expression in the tapetum. Our results showed that the hypermethylated cytosine located at the -37 position of the SCR-22 5' region of the S-22/S-44 heterozygotes or the -350 to -440 region of the SCR-22 5' region with a low methylation rate in S-22/S-36 heterozygotes is far from the core region for expression in the tapetum. Although a putative core binding sequence, CA(A/C)G(T/C)(T/C/A) (T/C/A), for a class of plant-specific NAC transcription factors was suggested to be present within the 5' region (nucleotides -37 to -31) of SCR-22 by a survey of putative cis-regulatory elements in silico 29 (Supplementary Figure S4), this putative cis-regulatory element was not found within the promoter region of SCR-47. Thus, the binding sequence of these NAC transcription factors is not essential for the expression of all class-I SCR/SP11 alleles in the tapetum. In addition, no methylated cytosine was detected in the class-I SCR-22 in microspores of S-22/S-44 or S-22/S-36 heterozygotes, suggesting that the suppression of SCR-22 expression is not induced by DNA methylation. Thus, the recessive SCR-22 is considered to be suppressed through a mechanism without a DNA methylation pathway in general. The hypermethylated cytosine in the SCR-22 promoter in the tapetum of S-22/S-44 heterozygotes or the cytosine with a low methylation rate in the -350 to -440 region of the SCR-22 5' region in the tapetum of S-22/S-36 heterozygotes may be the result of histone modification 22 . These findings suggest that the suppression of transcription of SCR-22 is not caused by the DNA methylation-mediated suppression through preventing the transcription factors from binding to their target sequence but possibly by the alteration of chromatin structure 22 . Therefore, a suppression mechanism different from that for class-II SCR/SP11 may function in the suppression of recessive class-I SCR-22 in S heterozygotes.
Class-I SCR/SP11 alleles have been reported to be transcribed in both the tapetum and microspores 7,10,16 . The transcription of class-I SCR/SP11 alleles in microspores occurs slightly later than that in the tapetum 12 . We found that SCR-22 was expressed in microspores of S-22 homozygotes but not in those of S-22/S-44 and S-22/S-36 heterozygotes. Although S-44 and S-36 were not present in the microspores having S-22 in the S-22/S-44 and S-22/S-36 heterozygotes, the expression of SCR-22 was suppressed, indicating that suppression of SCR-22 expression may have been induced to occur before meiosis by a mechanism different from that caused by Smi. Further analyses are required to elucidate the suppression mechanism of recessive class-I SCR/SP11 alleles in S heterozygotes.

Pollination tests
Pollinated flowers were placed on solid agar for 24 h at 21°C. Pistils were softened in 1 N NaOH at 55°C for 1 h. The pistils were then stained with 0.1% aniline blue in 0.1 M K 3 PO 4 and mounted in 60% glycerol. Pollen tubes were observed under a fluorescence microscope. Three flowers were used for each pollination, and the tests were replicated three times on different days.
Amplification of SCR-22 from B. rapa Total RNA was extracted from anthers of S-22 homozygotes using TRIzol reagent (Invitrogen, Shanghai, China). RNA was reverse-transcribed using a Super-Script™ III First-Strand Synthesis System (Invitrogen, Shanghai, China). A partial sequence of SCR-22 was amplified by nested PCR using the primers SP11-1 31 , SP11-F1 6 , SP11-Fa, and SP11-1F8 as forward primers and Not1-(dT)18 as a reverse primer for the first PCR and SP11-2 31 , SP11-F2 6 , and SP11-2Fa as forward primers and RT1-long 9 as a reverse primer for the second PCR. When SP11-1 and SP11-F2 were used for the first and second PCRs, a partial fragment of SCR-22 was amplified. The promoter region and entire DNA sequence of SCR-22 were identified by inverse PCR 32 . The primer sequences are listed in Supplementary Table S1.

Tapetum isolation
Thirty flowers were collected 3 days before anthesis for collecting anthers. Both ends of the anthers were cut away, and the remaining part of the anthers was cut into two equal sizes. The two pieces were then placed into 1 mL tapetum isolation buffer (50 mM Hepes buffer, 0.5 M sucrose, and KOH to adjust the pH value to 7.5) in 1.5 mL centrifuge tubes. The tube was vortexed for 30 min to release microspores. The solution containing microspores was removed by filtration using nylon net (0.5 mm pore size) and washed three times with tapetum isolation buffer. The solution was centrifuged at 100 × g for 10 min, and the pellet contained microspores. Then, 1 mL isolation buffer (5% cellulase RS and 1% pectolyase Y23 in tapetum cell isolation buffer) was added to the anthers and vortexed for approximately 7 min. The upper solution was transferred to a new tube by filtration using a nylon net (0.5 mm pore size), and the upper solution was centrifuged at 20,000 × g for 10 min. The pellet contained the tapetum. Total RNA and DNA were isolated from the tapetum using TRIzol reagent (Invitrogen, Shanghai, China) for the following experiments.
To examine the adequacy of our method, the anthers collected before and after the isolation process were embedded in 5% agar. The anthers were sliced into 30 μM-thin sections by a DTK-3000 microslicer (Dosaka, Kyoto, Japan), and the sections were observed by a microscope. RNA was reverse-transcribed by the SuperScript™ III First-Strand Synthesis System (Invitrogen, Shanghai, China). Real-time RT-PCR was performed using SsoAdvanced TM SYBR ® Green Supermix (Bio-Rad, Shanghai, China) on a Bio-Rad ® CFX96 system, following the manufacturer's instructions. Each SCR/SP11 region was amplified with specific primers (Supplementary Table S1). The Actin gene was used as an endogenous reference gene. The primers were confirmed to be approximately 90% to 100% efficient for amplification, and the 2 −ΔΔCT method 33 was used for all analyses. All reactions were performed in triplicate, and an average value was calculated for each set of reactions.

DNA methylation state detection
DNA was isolated from the tapeta of S-22 and S-44 homozygotes and S-22/S-36, S-22/S-44, and S-8/S-44 heterozygotes. The DNA was bisulfite treated with a MethylCode™ Bisulfite Conversion Kit (Applied Biosystems, Shanghai, China). The SCR-44 promoter region and the promoter region, coding region, and intronic region of SCR-22 modified by bisulfite were amplified using specific primers (Supplementary Table S1). Amplified PCR products were cloned into pGEM-T Easy vectors (Promega, Beijing, China), and at least 30 clones were sequenced.

Detection of mature and precursor Smi sRNA
Detection of mature Smi sRNA was performed as previously described 34 . When microspores were in the uninucleate stage, small RNA was isolated from the anthers of S-  Table S1). The transcribed products were quantified using SsoAdvanced TM SYBR ® Green Supermix (Bio-Rad, Shanghai, China) with small RNA-specific primers and universal primers (Supplementary Table S1). U6 was used as an endogenous reference gene. The 2 −ΔΔCT method 33 was used for all analyses. All reactions were performed in triplicate, and an average value was calculated for each set of reactions.
For detection of precursor Smi sRNA, total RNA was isolated from anthers when microspores were at the uninucleate stage. RNA was reverse-transcribed by the SuperScript™ III First-Strand Synthesis System (Invitrogen, Shanghai, China) with precursor Smi-22-specific primers and Actin-R (Supplementary Table S1). Real-time RT-PCR was performed using SsoAdvanced TM SYBR ® Green Supermix (Bio-Rad, Shanghai, China) on a Bio-Rad ® CFX96 system, following the manufacturer's instructions. The Actin gene was used as an endogenous reference gene. The primers were confirmed to be approximately 90% to 100% efficient for amplification, and the 2 −ΔΔCT method 33 was used for all analyses.