Dicer-like 3 produces transposable element-associated 24-nt siRNAs that control agricultural traits in rice

Significance

The functional relationship of transposons and small RNAs remains an important question in the study of gene expression and its effect on agronomic traits. Here, we use deep sequencing of small RNAs to provide the first evidence that the rice Dicer-like 3 homolog OsDCL3a produces 24-nt small interfering RNAs (siRNAs) predominantly associated with miniature inverted repeat transposable elements (MITEs). These 24-nt siRNAs target genes adjacent to MITEs and act as broadly functioning regulators of gene expression. In particular, OsDCL3a directly targets genes involved in homeostasis of the plant hormones gibberellin and brassinosteroid, thus controlling important agricultural traits. This mechanism of fine-tuning gene expression mediated by MITEs may be conserved in organisms with genomes rich in dispersed repeats or transposable elements.

Abstract

Transposable elements (TEs) and repetitive sequences make up over 35% of the rice (Oryza sativa) genome. The host regulates the activity of different TEs by different epigenetic mechanisms, including DNA methylation, histone H3K9 methylation, and histone H3K4 demethylation. TEs can also affect the expression of host genes. For example, miniature inverted repeat TEs (MITEs), dispersed high copy-number DNA TEs, can influence the expression of nearby genes. In plants, 24-nt small interfering RNAs (siRNAs) are mainly derived from repeats and TEs. However, the extent to which TEs, particularly MITEs associated with 24-nt siRNAs, affect gene expression remains elusive. Here, we show that the rice Dicer-like 3 homolog OsDCL3a is primarily responsible for 24-nt siRNA processing. Impairing OsDCL3a expression by RNA interference caused phenotypes affecting important agricultural traits; these phenotypes include dwarfism, larger flag leaf angle, and fewer secondary branches. We used small RNA deep sequencing to identify 535,054 24-nt siRNA clusters. Of these clusters, ∼82% were OsDCL3a-dependent and showed significant enrichment of MITEs. Reduction of OsDCL3a function reduced the 24-nt siRNAs predominantly from MITEs and elevated expression of nearby genes. OsDCL3a directly targets genes involved in gibberellin and brassinosteroid homeostasis; OsDCL3a deficiency may affect these genes, thus causing the phenotypes of dwarfism and enlarged flag leaf angle. Our work identifies OsDCL3a-dependent 24-nt siRNAs derived from MITEs as broadly functioning regulators for fine-tuning gene expression, which may reflect a conserved epigenetic mechanism in higher plants with genomes rich in dispersed repeats or TEs.

Transposable elements (TEs) and repetitive sequences make up more than 35% of the rice genome (1). TEs include DNA transposons, which mobilize by a “cut-and-paste” mechanism, and retrotransposons, which mobilize by a “copy-and-paste” mechanism (2, 3). Miniature inverted repeat TEs (MITEs), widespread, short (less than 600 bp), nonautonomous DNA transposons, occur in many plant and animal genomes, including Arabidopsis, rice, sorghum, maize, Caenorhabditis elegans, and humans (2, 4). These elements contain short terminal inverted repeats and target-site preference (TA or TAA) for target site duplication (2). MITEs are the highest copy-number TEs in rice and are mainly dispersed in the chromosomal arms (i.e., in gene-rich euchromatic regions), especially in the vicinity of genes (5). Miniature Ping (mPing) was the first discovered active MITE from any organism and the first active DNA transposons from rice (6–8). The copy number of mPing is highly polymorphic among japonica and indica (the two rice subspecies), temperate and tropical japonica (two subgroups of japonica), as well as domesticated rice (Oryza sativa) and ancestral species (Oryza rufipogon) (6, 7). In some rice lines, mPing elements underwent a massive amplification, with ∼40 new insertions per generation (9). These new mPing elements preferentially inserted into the 5′ flanking region of genes and avoided exons (10). These mPing elements can either regulate or have no detectable impact on the expression of nearby genes (10). It is proposed that the rapid and recent amplification of a subset of mPing confers nearby gene induction upon different stress conditions, implying that new stress-inducible alleles may generate in cultivated populations (10). Rice TEs also include about 14% long terminal repeat (LTR) retrotransposons and 1% non-LTR retrotransposons, which can be further classified as long interspersed elements (LINEs) and short interspersed elements (SINEs) (1). Several LTR retrotransposons, including Tos10, Tos17, and Tos19 (11–13), and a non-LTR retrotransposon LINE Karma, have been shown to be autonomous and active in rice (14, 15).

The mobilization of TEs provides a major driving force for gene and genome evolution (16). However, TE mobilization can also disrupt genome stability and most organisms have evolved diverse epigenetic mechanisms to repress TE activity. For example, epigenetic mechanisms, including DNA methylation and dynamic histone methylation, control transposon silencing (17, 18). In addition, distinct epigenetic mechanisms regulate different TEs (19).

Plants produce 24-nt small interfering RNAs (siRNAs) from repeats and TEs. These 24-nt siRNAs trigger DNA methylation at all CG, CHG, and CHH (where H = A, T, or C) sites, resulting in H3K9me2 modification. This modification reinforces transcriptional silencing of TEs and genes that harbor or are adjacent to repeats or TEs in Arabidopsis (20–25). In Arabidopsis, the plant-specific RNA polymerase IV (Pol IV) transcribes heterochromatic regions. RNA-dependent RNA polymerase 2 (RDR2) then synthesizes double-stranded RNA intermediates as precursors for RNase III-class Dicer-like 3 (DCL3) to process into 24-nt siRNAs (26). The siRNAs load into ARGONAUTE 4 (AGO4) and promote heterochromatin formation by DNA and histone methylation at the source loci (26). In Arabidopsis, repeats and TEs primarily occur around the centromeres or knob regions; in contrast, the rice genome consists of over 40% heterochromatin, which occurs in discontinuous and less distinct patterns (27, 28). Therefore, in rice more protein coding genes are exposed to repetitive sequences than in Arabidopsis (29). Rice contains two homologs of Arabidopsis DCL3, OsDCL3a, and OsDCL3b. OsDCL3b produces stamen-specific 24-nt phased small RNAs (30) and OsDCL3a produces 24-nt centromere-associated OsCentO siRNAs, MITE-derived siRNAs for abiotic stress responses, and noncanonical long miRNAs (lmiRNAs) (30–32). These lmiRNAs load into rice AGO4 homologs, and can direct DNA methylation at their target genes for transcriptional gene silencing (31). OsDCL3a activity is required for the biogenesis of miR820, which locates within CACTA DNA transposon and targets OsDRM2, the de novo DNA methyltransferase homolog in rice (23, 31, 33, 34). Therefore, OsDCL3a-dependent miR820 produced from a transposon suppresses the host’s silencing machinery (33).

High copy-number MITEs are dispersed throughout the rice genome. How the mechanisms regulating MITEs, such as 24-nt siRNAs associated with MITEs, affect gene expression, and thereby contribute to agricultural or developmental traits, remains unclear. Here, we show that OsDCL3a is primarily responsible for 24-nt siRNA processing in rice. Reduction of OsDCL3a expression levels by RNAi causes a genome-wide reduction of siRNAs and derepresses the heterochromatin status of MITEs, resulting in increased expression of nearby genes. OsDCL3a-dependent 24-nt siRNAs directly target gibberellin (GA) and brassinosteroid (BR) homeostasis-related genes. Our findings thus reveal important roles for OsDCL3a in 24-nt siRNA biogenesis and maintenance of heterochromatin status of siRNA-associated MITEs; suppression of these MITEs influences expression of nearby genes and affects important agricultural traits in rice.

Results

Knockdown of OsDCL3a Affects Important Agricultural Traits in Rice.

We previously used RNAi to knock down OsDCL3a expression, and generated two independent OsDCL3a RNAi lines, 3a-3 and 3a-1 (30). These two lines, 3a-3 and 3a-1, affect phenotypes with a severity correlated with the knockdown level of OsDCL3a. Compared with WT (Nipponbare), 3a-3 and 3a-1 plants showed significantly reduced plant height at heading stage (Fig. 1 A and B) and increased bending angle of the lamina joint (Fig. 1C). The angle of the flag leaf increased four- to sixfold in 3a-3 and 3a-1 compared with WT (Fig. 1D). Furthermore, 3a-3 and 3a-1 plants had smaller panicles than WT, which manifested as reduced primary and secondary panicle branches (Fig. 1 E–G). These phenotypes of 3a-3 and 3a-1 plants resemble the reported phenotypes of ago4ab-1 and rdr2-2 RNAi lines (Fig. S1) (31).

Fig. 1.

OsDCL3a knockdown plants display pleiotropic phenotypes affecting important agricultural traits. (A and B) OsDCL3a RNAi lines (3a-3 and 3a-1) show dwarf phenotypes (A) and statistical analysis of plant height (n = 30) (B). (C and D) Flag leaf inclination of WT, 3a-3, and 3a-1 (C) and statistical analysis of leaf angles (n = 30) (D). (E–G) Panicle morphology of WT, 3a-3, and 3a-1 (E) and statistical analysis of length (F) and branches of main panicle (G) (n = 30). **P < 0.01 with t test. Error bars correspond to the SD of biological repeats. (Scale bars: 10 cm in A and E; 2 cm in C.)

OsDCL3a-Dependent siRNAs Are Enriched in MITEs.

OsDCL3a produces 24-nt unphased, centromere-associated, MITE-derived siRNAs and lmiRNAs (30–32). To provide a genome-wide view of OsDCL3a-dependent 24-nt small RNAs (sRNAs), we compared small RNA accumulation in WT, 3a-3, and 3a-1 from the lamina joints of the flag leaf (Fig. S2A). We found that in the 3a-3 and 3a-1 lines, ∼82% (438,627 of 535,054) of total 24-nt sRNA clusters were reduced by more than threefold, compared with WT (Fig. S2). Further analysis revealed that 76% (333,692 of 438,627) of the OsDCL3a-dependent 24-nt sRNA clusters derive from repeats (Fig. 2A). In comparison with the annotation from plant repeat databases (http://plantrepeats.plantbiology.msu.edu/about.html), we identified 304,353 repeat- based OsDCL3a-dependent 24-nt sRNA clusters. These clusters include: MITE DNA transposons (33%, 101,053), other DNA TEs (20%, 59,996), retrotransposons (40%, 122,746, including 18% LTR, 1% non-LTR, and 21% unclassified retrotransposons), and other repeats (7%, 20,558) (Fig. S3A). Thus, the OsDCL3a-dependent siRNAs are significantly enriched in MITEs compared with retrotransposons (P = 0.006964, Fisher exact test), and other DNA TEs (P = 0.04709, Fisher exact test).

Fig. 2.

Distribution of OsDCL3a-dependent 24-nt siRNAs and different classes of repetitive sequences relative to up-regulated genes. (A) Pie chart showing the distribution of OsDCL3a-dependent 24-nt sRNAs loci based on Rice Genome Annotation Project Release 7 and miRBase release 20 annotations. (B) Differentially expressed genes in 3a-3 and 3a-1. Blue dots indicate up-regulated genes and green dots indicate down-regulated genes. (C) Abundance of 24-nt sRNAs in 2-kb regions upstream and downstream of up-regulated genes, with each gene annotated from the TSS to the TTS in WT (blue), 3a-3 (green), and 3a-1 (red). (D) The frequencies of four classes of repetitive sequences, MITEs (dark red), other DNA TEs (pink), retrotransposons (dark green), and other repeats (light green), were plotted in ±2 kb and gene body from TSS to TTS.

Effects of OsDCL3a-Dependent siRNAs on Gene Expression.

To test the effects of OsDCL3a-dependent siRNAs on the expression of their target genes and TEs, we performed RNA-seq on WT, 3a-3, and 3a-1. The correlation of RNA-seq data from 3a-3 and 3a-1 was ∼0.95, indicating the reproducibility of two independent samples of RNA libraries (Fig. S3B). Among the 859 differentially expressed genes altered by 1.5-fold or more, 679 were up-regulated and 180 were down-regulated in both 3a-3 and 3a-1 plants (Fig. 2B). Furthermore, we mapped 24-nt siRNAs and repetitive sequences to gene bodies and found that 24-nt siRNAs are enriched in 2-kb regions upstream of transcriptional start sites (TSS) and downstream of transcriptional terminal sites (TTS), regardless of gene-expression levels (Fig. 2 C and D and Fig. S3 C–F). The abundances of 24-nt siRNAs in the 5′ and 3′ regions of all genes were significantly lower in 3a-3 and 3a-1 (Fig. 2C and Fig. S3 C and E). The distribution of MITEs, but not other DNA TEs, retrotransposons, or repeats, was highly correlated with OsDCL3a-dependent 24-nt siRNAs (Fig. 2 C and D and Fig. S3 C–F). We also noticed that the up-regulated genes are nearly four times more numerous than the down-regulated genes. These data suggest that impairing OsDCL3a causes widespread reduction of 24-nt siRNAs from MITEs and other TEs, which mainly result in the up-regulation of nearby gene expression.

OsDCL3a Targets Genome-Wide Gene Expression.

Because 24-nt siRNAs usually cause gene silencing, we further examined the functions of these putative OsDCL3a targets by Gene Ontology analysis of the genes up-regulated in the 3a-3 and 3a-1 lines. These genes were enriched in Gene Ontology groups for metabolic processes, including cell wall organization, cell wall macromolecule catabolic process, defense response, GA catabolic process, and BR homeostasis, among others (Fig. S4). Intriguingly, genes for GA catabolic process and BR homeostasis may be associated with the phenotypes of the OsDCL3a RNAi lines, such as dwarfism and enlarged leaf angle, which resemble the phenotypes of rice plants with reduced GA and excess BR (35–38).

GA Homeostasis-Related Genes as Direct Targets of 24-nt siRNAs.

GAs, diterpenoid plant hormones, control diverse plant developmental processes, including stem elongation, leaf expansion, seed germination, and flowering (39). In rice, two gene clusters encode the enzymes at the branch points between biosynthesis of GA and other labdane-related diterpenoids (Fig. S5); these gene clusters include CYP76M7 (cytochrome P450 76 monooxygenase 7) and OsKSL7 (kaurene synthase-like 7) in one cluster, and OsCPS4 (syn-CPP synthase 4) and CYP99A3 (cytochrome P450 A3) in the other cluster (40). In 3a-3 and 3a-1 lines, the transcript levels of CYP76M7, OsKSL7, and CYP99A3 increased with decreasing accumulation of 24-nt siRNAs from MITEs in the 5′ or intron regions of the corresponding genes (Fig. S6). Moreover, 3a-3 and 3a-1 plants produced En/Spm-like TE-derived 24-nt siRNAs in the 5′ region of OsCPS4; also, OsCPS4 mRNA levels increased threefold (Fig. S6B). Using bisulfite sequencing and ChIP followed by quantitative PCR (qPCR) assays, we found that DNA methylation of TE regions was not significantly changed, whereas H3K9me2 levels largely decreased (Fig. S6). These results indicate that an OsDCL3a deficiency and loss of 24-nt siRNAs causes the up-regulation of genes critical for diterpenoid biosynthesis, which may influence GA biogenesis and therefore reduce plant height.

In addition, Elongated Uppermost Internode (EUI), which encodes a GA deactivating enzyme, also harbors OsDCL3a-dependent 24-nt siRNAs from three MITEs in its introns (Fig. 3 A–C). EUI was up-regulated in 3a-3 and 3a-1 lines along with a decrease in 24-nt siRNAs (Fig. 3 B and D). Furthermore, we also found that DNA methylation was not significantly altered, whereas H3K9me2 levels decreased in 3a-3 and 3a-1 compared with WT (Fig. 3 E and F). Overexpression of EUI causes severe dwarfism in rice (35, 41) (Fig. S5); thus, the reduction of 24-nt siRNAs from MITEs may elevate EUI expression and contribute to the reduced height of OsDCL3a-deficient plants. Therefore, MITE-associated 24-nt siRNAs epigenetically regulate GA anabolism and catabolism-related genes, which may affect rice plant height. To prove this hypothesis, we applied exogenous GA₃ and found that the dwarf phenotype of knockdown OsDCL3a plants can be rescued, further confirm that OsDCL3a-dependent 24-nt siRNAs regulate GA homeostasis-related gene expression and plant height (Fig. S7 A and B).

Fig. 3.

EUI encodes a GA deactivating enzyme and is activated in OsDCL3a RNAi lines. (A) Schematic representation of EUI. The arrowhead indicates the transcription start site. Boxes indicate exons (black), introns (white), UTR regions (gray), and MITEs (yellow). sRNA probes (SP5–SP7), bisulfite sequencing regions (BSP5, BSP6), and the regions used for ChIP-qPCR (R5–R7) are indicated by black lines. (B) sRNA-seq and RNA-seq data for EUI are shown in WT, 3a-3, and 3a-1. (C and D) Small RNA blot and qPCR validate the sRNA-seq and RNA-seq data, respectively. U6 probe and 5S rRNA/tRNA stained with ethidium bromide were used as small RNA blot loading controls. eEF1α was used as an internal reference for qPCR. The level of EUI transcription also was detected in ago4ab-1 and rdr2-2 lines. Error bars correspond to the SD. (E) Bisulfite sequencing analysis of the DNA methylation level of two MITE regions. (F) ChIP-qPCR assay detects the chromatin states using anti-H3K9me2. Anti-H3 was used as an internal reference for ChIP-qPCR. Error bars correspond to the SD.

BR Biogenesis-Related Genes as Direct Targets of 24-nt siRNAs.

We also found several BR biogenesis pathway genes, including OsGSR1 (GAST family gene in rice 1) and OsBR6ox (a rice BR-6-oxidase gene), to be likely direct targets of OsDCL3a. BR and GA pathways function together to regulate many biological processes (39, 42). For example, OsGSR1 promotes BR synthesis and represses GA20-ox-2, a major enzyme for GA biogenesis in rice (37) (Fig. S5). OsBR6ox catalyzes the C-6 oxidation step in BR biosynthesis (43, 44) (Fig. S5). BR is positively correlated with leaf angle in rice (38, 45). For example, OsGSR1 RNAi plants show an erect-leaf phenotype similar to plants deficient in BR (37). In 3a-3 and 3a-1 lines, we found fewer 24-nt siRNAs from MITEs, unchanged DNA methylation and reduced H3K9me2 levels in the 5′ region of OsGSR1 and OsBR6ox, and elevated mRNA levels (Fig. S8). Consistent with the enlarged leaf angle, we propose that BR biogenesis-related genes are activated in 3a-3 and 3a-1 lines.

In addition, we also found that two major components of the RNA-directed DNA methylation (RdDM) pathway, namely AGO4a and AGO4b (AGO4ab) and RDR2, also affected GA and BR homeostasis-related genes. In the ago4ab-1 and rdr2-2 RNAi background (Fig. 3D and Figs. S6 and S8), all GA and BR homeostasis-related genes are up-regulated. This result could be caused by effects on 24-nt siRNA associated MITEs or other TEs, and affect plant height and leaf angle, two important agricultural traits in rice. OsDRM2 is a major component in RdDM and suppressed by OsDCL3a-dependent miR820 at transcriptional and posttranscriptional level (31, 33). We found that, consistent with previous reports that OsDCL3a processes miR820 biogenesis (Fig. S9A), OsDRM2 accumulated approximately twofold higher levels in 3a-3 and 3a-1 lines (Fig. S9B), which may increase de novo DNA methylation and counteract the effect of loss of OsDCL3a activity.

The 24-nt siRNAs Epigenetically Regulate Os08g19420 and Alter Leaf Inclination.

Next, we tested whether 24-nt siRNAs from repeats elevated the expression of nearby genes and affect rice development. We examined Os08g19420, whose promoter region contains a Ditto-like MITE and seven tandem repeats, as well as a 3′ end including LTR, En/Spm-like TEs, and a SINE element (Fig. 4A). In 3a-3 and 3a-1 lines, the 24-nt siRNAs at above TEs and repeats were strongly reduced and the expression level of Os08g19420 was higher than WT (Fig. 4 B–D). Similar results were also obtained in ago4ab-1 and rdr2-2, indicating that the RdDM pathway is important for protein coding gene expression in rice (Fig. 4D). To evaluate whether these 24-nt siRNAs could mediate DNA methylation and H3K9me2 to affect expression of Os08g19420, we performed bisulfite sequencing and found a reduction of CHH methylation within the tandem repeats in 3a-3 and 3a-1 lines (Fig. 4E). Moreover, we found that the H3K9me2 levels of seven regions of the promoter, gene body, and 3′ end decreased in 3a-3 and 3a-1 (Fig. 4F). The downstream gene Os08g19440 has no transcriptional signals from RNA-seq data (Fig. 4B). Thus, a decrease in 24-nt siRNAs originating from 5′ and 3′ ends of Os08g19420 occurs with a decrease in repressive chromatin markers and an increase in expression of nearby genes.

Fig. 4.

24-nt siRNAs from MITEs and tandem repeats regulate expression of a nearby gene, Os08g19420. (A) Schematic of Os08g19420 with putative MITE Ditto-like and tandem repeats indicated by yellow boxes and arrows within the white box, respectively. Green boxes indicate LTR, En/Spm-like TEs, and a SINE element from left to right. Boxes indicate exons (black), introns (white), and UTR regions (gray). Black lines show the positions of siRNA probes (SP10–SP14), bisulfite sequencing regions (BSP9, BSP10), and five ChIP-qPCR analysis regions (R11–R17). (B) Small RNA-seq and RNA-seq data are shown in the region of Os08g19420 from WT (blue), 3a-3 (green), and 3a-1 (red). (C) Detection of 24-nt siRNAs by small RNA blot in WT, 3a-3, and 3a-1. (D) qPCR validation of Os08g19420 expression in WT, 3a-3, 3a-1, ago4ab-1, and rdr2-2 plants were normalized using the signal from eEF1α gene. The average ± SD values from three biological repeats are shown. (E) Bisulfite sequencing analysis of the DNA methylation level of MITE and tandem repeats. (F) In WT, 3a-3, and 3a-1, the chromatin states are detected by anti-H3K9me2 ChIP-qPCR assays at seven different regions. Anti-H3 was used as internal reference. Error bars correspond to the SD.

Additionally, to test whether overexpression of Os08g19420 could account for the developmental phenotypes in 3a-3 and 3a-1 lines, we transformed the intact Os08g19420 ORF driven by the OsActint1 promoter into WT (Nipponbare) plants. Four independent transgenic lines with significantly increased Os08g19420 expression levels showed exaggerated flag leaf angle at heading stage, similar to OsDCL3a RNAi lines (Fig. S10 A–C). These results demonstrate that 24-nt siRNAs repress Os08g19420 to control flag leaf inclination in rice.

To dissect the interplay of 24-nt siRNA-mediated regulation and MITE evolution, and to examine how MITE variants differentially affect nearby gene expression, we investigated a MITE in the Os08g19420 promoter region, examining its effect on gene expression among four rice japonica accessions. We noticed that japonica accessions from two subgroups, including two temperate japonica (TEJ), and two tropical japonica (TRJ) accessions, had a MITE-associated polymorphism in the Os08g19420 promoter regions (Fig. S10D). Further analysis showed that MITE partial deletion might cause Os08g19420 up-regulation in TRJ, but not in TEJ accessions, although we cannot rule out the possibility that the up-regulation of Os08g19420 was due to different genetic backgrounds (Fig. S10E).

Discussion

Our data collectively illustrate the widespread effect of OsDCL3a-dependent 24-nt siRNAs predominantly associated with MITEs and other TEs on the expression of nearby genes. This effect also controls important agriculture traits in rice. Compared with WT, knockdown OsDCL3a plants demonstrate a reduction in 24-nt siRNAs from TEs (5′ end, intron, and 3′ end), to a level that triggers genome-wide transcriptional up-regulation. This process results in derepression of GA and BR homeostasis-related genes, potentially accounting for the alterations of important agricultural traits in the RNAi lines (Fig. 5).

Fig. 5.

OsDCL3a-dependent 24-nt siRNAs regulate nearby gene expression and control rice development. In WT plants (A), 24-nt siRNAs (red lines) produced from TEs and repetitive sequences (yellow triangles) target nearby genes, including genes involved in GA and BR homeostasis. Normal regulation of GA and BR produces normal morphology. OsDCL3a RNAi knockdown (KD) plants (B) produce fewer 24-nt siRNAs (red lines), causing increased expression of genes involved in GA and BR homeostasis. Perturbed regulation of GA and BR biosynthetic genes results in an imbalance of GA and BR and causes abnormal morphology.

In this work, we found that knockdown OsDCL3a caused a significant reduction in 24-nt siRNAs and the levels of repressive chromatin marks, such as histone H3K9me2, leading to the activation of expression of nearby genes (Figs. 3 and 4 and Figs. S6 and S8). In addition, knockdown of AGO4ab and RDR2, two major components of the RdDM pathway in rice, produced a similar phenotype to the OsDCL3a knockdown, including derepression of 24-nt siRNA targets (Figs. 3D and 4D and Figs. S1, S6, and S8). In Arabidopsis, 24-nt siRNAs can direct DNA methylation and lead to histone H3K9 dimethylation to maintain TE silencing (46). Therefore, epigenetic regulation mediated by the small RNA pathway is conserved between rice and Arabidopsis. In addition, 24-nt siRNAs directly target many genes, including genes involved in plant hormone homeostasis, indicating that this regulatory mechanism plays broad roles in rice development.

We also noticed that the reduction of CHH methylation was only observed in the tandem repeats of the Os08g19420 promoter in 3a-3 and 3a-1 (Fig. 4E). In other MITEs, impaired production of 24-nt siRNAs did not reduce DNA methylation at CG, CHG, and CHH sites (Figs. 3E and 4E and Figs. S6 and S8). This result could be because of the reduced level of OsDCL3a-dependent miR820, which targets the major de novo methyltransferase, OsDRM2 (31, 34). Indeed, we observed a reduction of miR820 levels and an increase of OsDRM2 levels in 3a-3 and 3a-1 lines, which may promote de novo DNA methylation and counteract the effect of impaired OsDCL3a activity (Fig. S9). If increased, OsDRM2 activities override the effect of OsDCL3a, which will lead to decreased or unchanged nearby gene expression (Fig. S3 C–F). Alternatively, in addition to the RdDM pathway, CHH methylation at long TEs can be mediated in parallel by the nucleosome remodeler DDM1 (Decrease in DNA Methylation1) and another DNA methyltransferase, chromomethylases 2 (CMT2) in Arabidopsis (19). The rice CMT2 homolog (Os05g13780) also exists (19), which may be responsible for MITE-associated CHH methylation in OsDCL3a RNAi lines.

In contrast to rice, in which 24-nt siRNAs affect genome-wide gene expression, in Arabidopsis only a few genes (e.g., FWA, SDC, FLC, and RPP7) have been shown to be regulated by nearby TEs or repeats (20–22, 25, 47). In addition, unlike knockdowns of OsDCL3a, AGO4a4b, RDR2, and DRM2 in rice, which display pleiotropic developmental defects, Arabidopsis mutants in genes in the RdDM pathway, such as PolIV, RDR2, and DCL3, do not display obvious developmental phenotypes (23, 48, 49). The different effects on development in the two species may relate to the presence of more TEs (like MITEs) or repeats near genes in a complex genome, such as rice. Indeed, in crops such as maize, with complex TE- and repeat-rich genomes, RdDM pathway mutants also show pleiotropic developmental defects (50, 51).

MITEs, the most abundant DNA transposons, are the ultimate genomic parasites and occur very close to genes in rice (5, 52). This raises the question of how MITE insertions affect the behavior of neighboring genes. In this report, we experimentally demonstrated that in rice, OsDCL3a-dependent 24-nt siRNAs are substantially associated with MITEs or other TEs, and to some extent negatively regulate the expression of nearby genes (Figs. 2–4 and Figs. S6 and S8), which is consistent with a recent bioinformatic analysis (53). In addition, MITEs harbor regulatory sequences that may also act as enhancers to up-regulate expression of nearby genes (10, 54). Hence, MITEs might play a dual role in regulation of nearby genes, epigenetically repressing and genetically enhancing gene expression.

The polymorphisms in MITE sequences positively correlate with the variation in gene expression (Fig. S10 D and E). In the buttercup family (Ranunculaceae), an MITE inserted into an intron of the petal identity gene APETALA3-3 (AP3-3) results in gene silencing and petals transformed into sepals in apetalous Nigella (55). Thus, a burst of MITE activity, as has been observed in some rice strains (10), could generate epialleles for important agricultural traits. These epialleles could be selected and potentially adopted during domestication (56). In rice, epigenetic silencing of the DWARF1 (Epi-d1) cause a metastable dwarf phenotype, whereas Epi-df is a gain-of-function of FERTILIZATION-INDEPENDENT ENDOSPERM 1(FIE 1) epiallele, which shows pleiotropic defects (57, 58). We hypothesize that epialleles might be widespread and important for agricultural traits in rice.

Thus, our results provide evidence showing that 24-nt siRNA-associated MITEs and other TEs globally affect expression of nearby genes and control agricultural traits in rice. With advances in epigenomics and phenomics, it possible to enhance our ability to determine how often epigenetic state changes caused by TEs have been selected for agricultural traits during plant evolution, particularly in crops.

Materials and Methods

The OsDCL3a (3a-1, 3a-3), AGO4a4b and RDR2 RNAi lines, as well as TEJ1 (Nipponbare), TEJ2 (IRGC 418), TRJ1 (IRGC 17757), and TRJ2 (IRGC 328), were obtained from previous studies (30, 31, 59). Small RNA-seq and RNA-seq were analyzed and validated in WT, 3a-3, and 3a-1. Bisulfite sequencing and ChIP-qPCR assay was used to detect the levels of DNA and histone methylation. Details of experimental procedures are described in SI Materials and Methods. See Dataset S1 for the primers used in this study.