The RNA Export Factor ALY1 Enables Genome-Wide RNA-Directed DNA Methylation

The mRNA export protein ALY1 enables genome-wide RNA-directed DNA methylation, transposon repression, and transgene silencing.

Abstract

RNA-directed DNA methylation (RdDM) is a set of mechanisms by which transcriptionally repressive DNA and histone methylation are targeted to viruses, transposable elements, and some transgenes. We identified an Arabidopsis (Arabidopsis thaliana) mutant in which all forms of RdDM are deficient, leading to transcriptional activation of some transposable elements and the inability to initiate transgene silencing. The corresponding gene, ALY1, encodes an RNA binding nuclear export protein. Arabidopsis ALY proteins function together to export many messenger RNAs (mRNAs), but we found that ALY1 is unique among this family for its ability to enable RdDM. Through the identification of ALY1 direct targets via RNA immunoprecipitation sequencing, coupled with mRNA sequencing of nuclear and cytoplasmic fractions, we identified mRNAs of known RdDM factors that fail to efficiently export from the nucleus in aly1 mutants. We found that loss of RdDM in aly1 is a result of deficient nuclear export of the ARGONAUTE6 mRNA and subsequent decreases in ARGONAUTE6 protein, a key effector of RdDM. One aly1 allele was more severe due to an additional loss of RNA Polymerase V function, which is also necessary for RdDM. Together, our data reconcile the broad role of ALY1 in mRNA export with the specific loss of RdDM through the activities of ARGONAUTE6 and RNA Polymerase V.

INTRODUCTION

Small RNAs direct chromatin modifications in fungi, plants, and animals (reviewed in Castel and Martienssen, 2013). In plants, the RNA-directed DNA methylation (RdDM) pathway functions to initiate transcriptional gene silencing of transposable elements (TEs) and some transgenes (Chan et al., 2004; Teixeira et al., 2009), guiding the chromatin marks that delineate the boundary between heterochromatic and euchromatic genic regions (Li et al., 2015). RdDM is the only known mechanism by which cytosine DNA methylation and the transcriptionally repressive modification histone H3 Lys 9 dimethylation (H3K9me2) are de novo targeted in plants. Once these chromatin marks are established, they can be propagated without the requirement for RdDM via positive feedback loops of DNA methylation and H3K9me2 (reviewed in Zhang et al., 2018). This results in transcriptional silencing that is epigenetically maintained across generations (reviewed in Quadrana and Colot, 2016). In plant genomes, full-length TEs are often found in this heterochromatic maintenance silencing phase, where their transcriptional repression is only dependent on accurate replication of DNA and histone methylation patterns (Panda et al., 2016). By contrast, small, fragmented, and/or evolutionarily young TEs near genes are on the boundary of euchromatin and therefore require the constant reestablishment of chromatin marks by RdDM to repress their transcription (Zhong et al., 2012; Zemach et al., 2013).

The mechanisms of RdDM, as dissected in the reference plant Arabidopsis (Arabidopsis thaliana), involve an upstream small interfering RNA (siRNA)-generating phase and a downstream targeting of chromatin phase that is dependent on the generation of scaffolding transcripts, which remain tethered to the loci that produced them (reviewed in Zhang et al., 2018). These scaffolding transcripts are generated by RNA Polymerase V (Pol V), a plant-specific additional version of Pol II that has subfunctionalized to generate the scaffolding transcripts for RdDM (Kanno et al., 2005; Pontier et al., 2005). Pol V requires the factors DRD1 and KTF1, which act to facilitate Pol V scaffolding transcript production (Wierzbicki et al., 2008; He et al., 2009). In the upstream phase, the siRNAs that participate in RdDM are mostly generated by transcription of RNA Pol IV, another subfunctionalized plant-specific version of Pol II (Onodera et al., 2005). Pol IV transcripts are made doublestranded by RNA-dependent RNA Polymerase 2 (RDR2) and subsequently processed into siRNAs that are 24 nucleotides in length (Xie et al., 2004). These siRNAs are loaded into the ARGONAUTE 4 (AGO4) or AGO6 proteins (Havecker et al., 2010), which target Pol V scaffolding transcripts in the nucleus, triggering de novo chromatin modifications (Wierzbicki et al., 2009; Liu et al., 2018). In addition to the function of Pol IV, Pol II transcripts can also generate siRNAs for RdDM (21- to 22-nucleotide siRNAs in this case) that can be loaded into AGO6 to target chromatin modification (McCue et al., 2015). Our previous work demonstrated that AGO6 is the key protein necessary for the de novo targeting of chromatin modifications to new or active TEs (McCue et al., 2015).

In contrast to the processing of TE RNAs into siRNAs, genic messenger RNA (mRNA) transcripts are translated once exported from the nucleus. Transcription, mRNA processing, and nuclear export of genic mRNAs relies on the TRanscription-EXport (TREX) complex (Katahira, 2012). The TREX complex was first identified in yeast (Saccharomyces cerevisiae) and consists of the THO core complex (Hpr1, Tho2, Mft1, and Thp2) that is responsible for coupling transcription to mRNA processing. In yeast, TREX is recruited cotranscriptionally to Pol II–transcribed genes and plays a role in transcript elongation and mRNA nuclear export (Strässer et al., 2002). In addition to the THO core, TREX complex RNA binding proteins Tex1, Sub2, and Yra1 are required for nuclear mRNA export. In metazoans, TREX is recruited to mRNAs in a capping- and splicing-dependent manner (Rehwinkel et al., 2004; Masuda et al., 2005; Cheng et al., 2006). Less is known about the TREX complex in plants, but investigation of the highly conserved TREX component TEX1 in Arabidopsis has revealed that plant THO/TREX composition closely resembles the yeast and metazoan complexes (Yelina et al., 2010; Sørensen et al., 2017).

Once genic mRNAs have been transcribed and processed, the TREX component RNA binding protein Yra1 (Aly/REF in metazoans, ALY in plants) is recruited to the mRNA to facilitate mRNA nuclear export (reviewed in Cullen, 2003). Four ALY proteins exist in Arabidopsis (ALY1 to ALY4; Pfaff et al., 2018). Proteins in this family contain RNA recognition motifs, and in metazoans and yeast their recruitment to mRNAs is dependent on interaction with the RNA helicase UAP56 (Sub2 in yeast; Luo et al., 2001; Strässer and Hurt, 2001). UAP56 itself is thought to be recruited to mRNAs via multiple mechanisms including splicing, mRNA structure, transcription elongation, and 3′ end formation (Cullen, 2003). Once bound to an mRNA, ALY/REF interacts with the TAP-p15 transport receptor heterodimer (Mex67-Mtr2 in yeast), which facilitates Ran-independent nuclear export of the bound mRNA (Taniguchi and Ohno, 2008).

In addition to the role of Arabidopsis TREX in genic mRNA processing and export, multiple studies have implicated mRNA export factors in the production of microRNAs and siRNAs. While the mechanism is still unclear, the TREX complex may play a role in the transport of siRNA precursor RNAs (such as TE RNAs) to the appropriate processing machinery (Jauvion et al., 2010; Yelina et al., 2010; Francisco-Mangilet et al., 2015).

To explore how the processing of genic and TE RNA differs, we performed a reverse genetics screen for nuclear RNA binding proteins with dysfunctional RdDM. Mutations in the four members of the Arabidopsis ALY protein family were included in our screen because of their known targeting by the viral suppressor protein P19 (Uhrig et al., 2004), which targets and interferes with host RNA silencing to overcome viral repression (Silhavy et al., 2002). We found that one member of the ALY protein family, ALY1, is critical for RdDM and therefore enables the initiation of transgene silencing and TE repression.

RESULTS

ALY1 Is Necessary for RdDM

We utilized a reverse genetics approach to screen RNA binding proteins for defective RdDM by assaying the DNA methylation at multiple loci targeted by different RdDM pathways. We focused our analysis on mutants of Arabidopsis ALY1 (AT5G59950), which encodes an RNA binding protein highly similar to the metazoan ALY/REF family of well-studied mRNA nuclear export pathway proteins (reviewed in Cullen, 2003). Methylation at the SimpleHat2 TE, a canonical Pol IV-RdDM target (Pol IV-RdDM target = RdDM dependent on Pol IV-derived 24 nucleotide siRNAs; Lahmy et al., 2009), was absent or significantly reduced in three distinct alleles of aly1 (Figure 1A). As a negative control we used the ago6 mutant, which is necessary for RdDM in meristematic and floral tissues (Eun et al., 2011). To determine whether this loss of Pol IV-RdDM was locus specific or genome wide, we performed whole-genome cytosine methylome sequencing (MethylC-seq) of the aly1 mutant and corresponding controls (sequencing statistics shown in Supplemental Table 1). At TE loci targeted by Pol IV-RdDM, we identified a global loss of CHH (where H is any nucleotide except G) methylation similar to the pol V mutant (Figure 1B), in which RdDM collapses (Stroud et al., 2013), demonstrating that Pol IV-RdDM is dependent on ALY1. To assay the distinct Pol II expression–dependent RdDM pathway, we performed locus-specific DNA methylation analysis on the endogenous TAS3 locus, which is a target of RDR6-RdDM (RdDM that is dependent on RDR6’s production of 21- to 22-nucleotide siRNAs from a Pol II transcript; Panda et al., 2016). Two independent alleles of aly1 had absent or significantly reduced levels of TAS3 CHH methylation, while the third allele was reduced, but not to a statistically significant level (Figure 1C). We noted that the aly1-2 allele was consistently stronger (loses more methylation) compared with the other aly1 alleles for all loci tested. The dependence of RDR6-RdDM on ALY1 was also confirmed genome wide, where the CHH methylation level in the strong aly1-2 allele was as low as pol V at RDR6-RdDM–targeted TEs (Figure 1D). These results demonstrate that ALY1 is broadly necessary for the function of RdDM, but there are allele-dependent differences in the strength of the phenotype.

Figure 1.

ALY1 Is Necessary for RdDM.

(A) Locus-specific DNA methylation analysis of a Pol IV-RdDM target TE, SimpleHat2. Methylation is separated into the CG, CHG, and CHH sequence contexts (H=A,T,C), but only CHH context cytosines are present in this region of SimpleHat2. Error bars for DNA methylation data represent 95% Wilson confidence intervals. Asterisk (*) indicates a P-value of <0.05 determined by Student’s t test.

(B) Meta-analysis of CHH methylation at all Pol IV-RdDM targeted TEs in three biological replicates of the wild-type Col (1 to 3), aly1, and pol V. The number of TEs targeted by Pol IV-RdDM and assayed in this analysis is indicated in parentheses. For all meta-analyses, the solid line indicates the average methylation level and the transparent ribbon is the 95% confidence interval. wt Col, wild-type Columbia.

(C) Locus-specific DNA methylation analysis of TAS3, a RDR6-RdDM target region.

(D) Meta-analysis of CHH methylation at all RDR6-RdDM–targeted TEs.

(E) to (G). Meta-analysis of (E) CG methylation at all TEs with MET1-dependent CG context methylation, (F) CHG methylation at all TEs with CMT3-dependent CHG context methylation, and (G) CHH methylation at all TEs with CMT2-dependent CHH context methylation.

We next investigated whether ALY1 is required for maintenance of previously established DNA methylation patterns (in contrast to RdDM). As controls, we identified the DNA methylation levels of TEs wild-type Columbia (Col) and maintenance methylation factor mutants specific to each cytosine sequence context (CG=met1, CHG=cmt3, CHH=cmt2; reviewed in Zhang et al., 2018). We found that ALY1 was not responsible for the maintenance of CG methylation, while there was a slight reduction of maintenance CHG and CHH methylation (Figures 1E to 1G). The minor reduction of CHG and CHH methylation may be an indirect consequence of loss of a global RdDM, as it was similar to the slight reduction of CHG/CHH methylation found in the pol V mutant (Figures 1F and 1G). Therefore, we conclude that ALY1 is necessary for TE RdDM but functions minimally in the maintenance of DNA methylation.

To determine whether ALY1 enables RdDM at regions other than TEs, we calculated methylation levels at inverted repeats (IRs) genome wide. We found a loss of IR CHH methylation in aly1 similar to that of the pol V mutant (Supplemental Figure 1), demonstrating that all tested RdDM was deficient in aly1 mutants.

To examine whether loss of RdDM is a general feature of ALY family mutants, we screened aly2, aly3, and aly4 but found no reduction in RDR6-RdDM or Pol IV-RdDM (Supplemental Figure 2). We also found that loss of RDR6-RdDM is not a general feature of perturbed mRNA export, as RDR6-RdDM functioned correctly in other well-described mRNA export mutants tex1 and hpr1 (Supplemental Figure 3). Together, our data demonstrate that ALY1 is distinct among ALY protein family members and mRNA export proteins for its role in RdDM.

ALY1 Functions in TE Silencing

To identify the molecular consequence of ALY1 dysfunction on global mRNA levels, we performed mRNA-seq in three biological replicates of wild-type Col and aly1 (sequencing statistics in Supplemental Table 1). Using stringent cutoff criteria (log₂ fold change greater than ±1; false discovery rate [FDR] < 0.01), we found 10 genes with increased steady-state mRNA levels and 17 genes with decreased levels in aly1 (Figure 2A; genes listed in Supplemental Table 2). We confirmed that ALY1 itself had significantly reduced mRNA accumulation in the aly1 mutant. We additionally found two differentially expressed genes that are known to be responsive to the methylation of neighboring TEs: the expression of Repressor Of Silencing 1 decreases when RdDM function is lost (Williams et al., 2015), while the expression of SDC is increased upon loss of non-CG DNA methylation (Henderson and Jacobsen, 2008). We also identified a cluster of neighboring genes with reduced mRNA accumulation in aly1 (Figure 2A, blue labels). We explored the coverage of this locus in our MethylC-seq data (data not shown) and found a 12.6-kb deletion present in the aly1-2 mutant line encompassing the six genes AT4G22285 to AT4G22310 (gene annotations provided in Supplemental Figure 4A). Deletions can occur during the process of transgenesis, which was used to generate the aly1-2 line (Nacry et al., 1998; Jupe et al., 2019). We found that this deletion did not cause the loss of RdDM phenotype in aly1-2, as the deletion was not present in the aly1-1 mutant line (Supplemental Figure 4B), which displays decreased RdDM (Figures 1A and 1C). Additionally, after crossing and segregation in an F2 population, TE RdDM remained dysfunctional in the aly1-2 ddm1 double mutant that retained this cluster of genes (Supplemental Figures 4B and 4C).

Figure 2.

ALY1 Functions in the Suppression of TE Expression via RdDM.

(A) mRNA-seq analysis of gene expression in aly1 using stringent cutoff criteria (log₂ fold change greater than ±1.0 and FDR < 0.01). Dashed lines indicate a log₂ fold change of ±1.0. Blue gene name labels indicate the cluster of genes on chromosome 4 that are deleted in the aly1-2 mutant line.

(B) mRNA-seq analysis of genome-wide TE expression in aly1. The 19 TEs labeled in red are differentially expressed based on the stringent criteria.

(C) CHH methylation levels of the 19 differentially expressed TEs from (B) in the wild-type and mutant genotypes. The diamond represents the mean and the bars the 95% confidence interval. The three wild-type Col samples represent distinct biological replicates. wt Col, wild-type Columbia.

(D) Chromatin IP-qPCR of H3K9me2 at the centromeric TE family Athila6A and the intergenic TE SimpleHat2 in the indicated genotypes. Individual biological replicates are shown as dots, and the mean and se are indicated.

(E) Expression of Athila6A TEs pulled out of the larger TE analysis in (B). Box plots display the median normalized to total library size, box edges represent the first and third quartiles, and the whiskers extend to 1.5 times the interquartile range.

As predicted due to the loss of TE RdDM, we detected a small but measurable increase in TE expression in aly1. We found that 19 TEs had significantly increased TE mRNA levels (and 0 had decreased levels; Figure 2B; Supplemental Table 2), with an overrepresentation of the AtGP1 TE family (P = 8.1e−12). We confirmed that the DNA methylation of the 19 reactivated TEs was generally lost in the RdDM mutants aly1 and pol V (Figure 2C). For TEs that are controlled by maintenance methylation (and not RdDM), ALY1 dysfunction did not result in transcriptional reactivation. For example, the centromeric Athila6A TE family is targeted in wild-type by maintenance DNA methylation only (Panda et al., 2016) and therefore did not lose H3K9me2 in aly1 (Figure 2D) and was not transcriptionally activated in aly1 mutants (Figure 2E). This further demonstrates that ALY1 function is specific to RdDM and that it is not a general TE silencing factor. However, significantly more TEs lost methylation (8952) than were transcriptionally reactivated in aly1 (19). For example, SimpleHat2 lost DNA methylation due to the lack of RdDM in the aly1 mutants (Figure 1A), but SimpleHat2 was not transcriptionally reactivated (Figure 2B). Furthermore, SimpleHat2 H3K9me2 was lost in aly1 (Figure 2D), demonstrating that the deficiency in aly1 was strong enough to inhibit RdDM and H3K9me2 at this TE, but like many other RdDM mutants, this alone was not strong enough to result in the transcriptional reactivation of most TEs (Blevins et al., 2014).

ALY1 Acts Downstream of siRNA Production

We investigated whether ALY1 regulates the upstream phase of RdDM and therefore results in defective RdDM due to the loss of siRNAs. We examined siRNA accumulation via three distinct assays. First, we examined the plant phenotype, as leaf narrowing occurs when TAS trans-acting siRNAs do not participate in posttranscriptional gene repression (Peragine et al., 2004). aly1 mutants did not show leaf narrowing similar to rdr6 mutants (Figure 3A), indicating that siRNA-driven posttranscriptional gene silencing functions correctly in aly1 mutants. However, aly1 mutants did display a wavy/wrinkled leaf blade (Figure 3A), which is accounted for by the overexpression of SDC (Figure 2A), which generates this phenotype when RdDM is lost (Henderson and Jacobsen, 2008). Furthermore, aly1 did not have the other phenotypic hallmarks associated with the loss of siRNA production (Supplemental Figure 5; Pfaff et al., 2018). Second, we used RNA gel-blot analysis to determine that TAS3 21- to 22-nucleotide siRNAs accumulated to the wild-type levels in the aly1 mutant (Figure 3B), even though RdDM of this locus was defective in aly1 (Figure 1C). Third, we performed small RNA deep sequencing (statistics in Supplemental Table 1) and found that the global accumulation of aly1 small RNAs was within the range of variability between different biological replicates of wild-type Col (Figure 3C). We investigated the 24-nucleotide siRNA accumulation from regions undergoing maintenance CHH methylation and found no loss of siRNA accumulation in aly1 (Figure 3D), which was confirmed by RNA gel-blot analysis of the Athila6A long terminal repeat (LTR; Figure 3E), demonstrating that ALY1 acts downstream of siRNA production. However, upon close inspection of the accumulation of 24-nucleotide siRNAs that specifically function in RdDM, we identified a decreased accumulation in aly1 mutants (Figure 3F). This reduction was confirmed for the SimpleHat2 target of Pol IV-RdDM (Figure 3E; Lahmy et al., 2009) and is similar to the 24-nucleotide siRNA reduction observed in pol V mutants (Figure 3F; Mosher et al., 2008). This loss of 24-nucleotide siRNAs in pol V is thought to be an indirect consequence of the global loss of RdDM-targeted H3K9me2 (Mosher et al., 2008), which signals the feedback-loop recruitment of Pol IV and continued production of 24-nucleotide siRNAs (Herr et al., 2005; Law et al., 2013). Therefore, we believe that the loss of some 24-nucleotide siRNAs in aly1 is likely due to this same indirect consequence of the global loss of RdDM.

Figure 3.

Small RNA Production in the aly1 Mutant.

(A) Leaf phenotypes of juvenile plants before flowering.

(B) Small RNA gel blot of TAS3-derived 21- to 22-nucleotide (nt) siRNAs. The blot was reprobed for tRNA^Met as a loading control.

(C) Total small RNA (microRNA + siRNA) accumulation. Three biological replicates of wild-type Col were used to generate sd bars for each size class. RPM, reads per million mapped; wt Col, wild-type Columbia.

(D) Accumulation of 24-nt siRNAs specifically from regions that undergo maintenance CHH methylation (cmt2 differentially methylated regions [DMRs]).

(E) Small RNA gel blot of nucleus- and cytoplasm-enriched cellular fractions probed multiple times.

(F) Accumulation of 24-nt siRNAs specifically from regions that undergo RdDM (pol V DMRs). Significance was determined by Student’s t test with Welch correction, P-value < 0.001.

Lastly, because ALY proteins are known to function in RNA export, we determined whether siRNA accumulation in the cytoplasm occurs normally in aly1 mutants. We used small RNA from nucleus- and cytoplasm-enriched fractions from the same cell extracts to determine that export of 24-nucleotide siRNAs occurred normally from loci that still produced siRNAs in aly1 (Athila6A; Figure 3E). Our data therefore demonstrate that ALY1 does not function in small RNA production or export, and the loss of siRNAs from RdDM loci in the aly1 mutant is likely an indirect effect of the global loss of RdDM akin to a pol V mutant.

Initiation of Transgene Silencing Is Defective in aly1 Mutants

We next aimed to determine whether the loss of siRNAs from RdDM regions is the cause or a consequence of RdDM failure in aly1. We used a transgene carrying a transcriptionally incompetent fragment of an Athila6A TE (Athila6A LTR Δpro), which is targeted in trans for RdDM via 24-nucleotide siRNAs produced from endogenous Athila6A TEs (Fultz and Slotkin, 2017). These 24-nucleotide siRNAs are present in aly1 mutants (Figure 3E) because these endogenous TEs undergo maintenance CHH methylation, not RdDM. We performed DNA methylation analysis in the first (T1) generation after the integration of the transgene and found that aly1 mutants were defective for homology-based silencing via 24-nucleotide siRNAs to a similar extent as pol IV mutants (Figure 4A), which lack the 24-nucleotide siRNAs from endogenous Athila TEs altogether (McCue et al., 2012). These data demonstrate that even when 24-nucleotide siRNA production occurs in aly1, RdDM is still defective, proving that the lack of siRNAs from endogenous RdDM loci in aly1 is a consequence of RdDM failure, and not the root cause of defective RdDM.

Figure 4.

ALY1 Is Necessary for the Initiation of Transgene Silencing.

(A) Homology-based silencing via 24-nucleotide (nt) siRNAs assayed using the Athila LTR Δpro transgene. Error bars for DNA methylation data represent 95% Wilson confidence intervals.

(B) Homology-based silencing via 21- to 22-nt siRNAs assayed using the TAS3 internal region transgene. CHG context DNA methylation is highly variable due to the low number of CHG context sites within this PCR region.

(C) Expression-dependent silencing assayed using the exogenous iTto tobacco TE transgene. All transgenes were previously described (Fultz and Slotkin, 2017).

To further test ALY1’s role in the establishment of transgene silencing, we tested two additional transgene systems that specifically assay unique mechanisms of silencing initiation (Fultz and Slotkin, 2017). We tested homology-based silencing targeted by 21- to 22-nucleotide siRNAs by integrating a transcriptionally incompetent fragment of the TAS3 gene (TAS3 internal region), which is targeted for RdDM by 21- to 22-nucleotide siRNAs produced from the endogenous TAS3 locus (Panda et al., 2016). aly1 mutants were defective for homology-based silencing via 21- to 22-nucleotide siRNAs to a similar extent as the rdr6 mutant (Figure 4B), which lacks 21- to 22-nucleotide siRNAs from the endogenous TAS3 locus (Allen et al., 2005). We also tested expression-dependent silencing, which is not based on the transgene’s identity to the genome, by integrating an expressed exogenous tobacco TE (iTto) (Fultz and Slotkin, 2017). The aly1 mutants were deficient for expression-dependent initiation of transgene silencing to a similar extent as pol V mutants (Figure 4C). Together, our data demonstrate that ALY1 is necessary for initiation of all known forms of transgene silencing.

ALY1 Interacts with a Broad Array of mRNAs

To pinpoint what aspect of RdDM is perturbed in aly1 mutants, we performed an RNA immunoprecipitation (RIP) using an ALY1pro:ALY1-GFP (green fluorescent protein) transgene (ALY1-GFP; Pfaff et al., 2018), which partially complemented the loss of RdDM in the aly1-1 mutant (Figure 5A). We performed RIP-seq of ALY1-GFP in the aly1-1 background in four replicates and two mock immunoprecipitations (IPs; Supplemental Data Set 1). Since the aly1-2 loss of RdDM was similar to that of the pol V mutant (Figures 1B and 1D), we first tested whether ALY1-GFP protein interacts with Pol IV or Pol V transcripts. We failed to detect enrichment of those RNAs (Figure 5B); of the regions of the genome that undergo RdDM (1352 Pol V hypo-differentially methylated regions [DMRs], data from Figure 3F), only 649 had at least one read in any of the RIP-seq libraries, and of these only 1 was statistically enriched in the ALY-GFP RIP-seq (Figure 5B), suggesting that ALY1 does not bind Pol IV- or Pol V-generated RNAs. We instead determined that, as expected, ALY1 binds genic mRNA transcripts: the ALY1-GFP RIP-seq–enriched 5446 genic mRNAs, including 325 above a stringent threshold (log₂ fold change greater than ±1; FDR < 0.01; Figure 5C). Since there are 16,360 genes expressed in the wild-type inflorescence tissue (threshold >50 normalized counts, data extrapolated from Figure 2A), this suggests that ALY1 directly interacts with only 2 to 33% of expressed mRNAs. The 325 enriched mRNAs were expressed in RNA-seq from the same tissue type (data from Figure 2) as expected, but did not represent the highest expressed genes in that tissue (Supplemental Figure 6A), demonstrating the specificity of the RIP enrichment.

Figure 5.

Identification of ALY1 Direct Targets by RIP-Seq.

(A) Partial complementation of the methylation defect of the aly1-1 allele using the ALY1-GFP transgene. The wild-type Col and aly1 data are from Figures 1A and 1C. wt Col, wild-type Columbia.

(B) ALY1-GFP RIP-seq analysis of RNAs produced from regions of the genome undergoing RdDM (pol V DMRs). The number of genes represented in each section of the volcano plot is listed above.

(C) RIP-seq analysis of genic mRNAs. Genes with known or suspected roles in small RNA biogenesis, RdDM, and chromatin modification are labeled.

(D) Fluorescence in situ hybridization detection of polyadenylated mRNAs in the wild-type and aly1 developing roots.

(E) GO enrichment analysis of the 325 statistically enriched mRNAs in the ALY1-GFP RIP-seq.

To validate our finding that ALY1 does not function on all mRNAs as a general mRNA nuclear export factor, we performed an in situ localization of polyadenylated mRNA in the aly1 mutant. In the Arabidopsis aly1 aly2 aly3 aly4 quadruple mutant (4x aly), mRNA is retained in the nucleus on a global level (Pfaff et al., 2018). By contrast, we found the same qualitative cellular distribution of mRNA in the aly1 single mutant as in wild-type Col (Figure 5D). Polyadenylated mRNA was concentrated in the cytoplasm and nuclear periphery, but not localized diffusely in the nucleoplasm as it is in the 4x aly mutant (Pfaff et al., 2018). These data support previous findings that the full role of the metazoan ALY/REF protein in mRNA export has been subdivided in plants between ALY1, ALY2, ALY3, and ALY4 (Pfaff et al., 2018).

Among the ALY1-GFP RIP-enriched genes were several that regulate small RNA production and chromatin modification (labeled mRNAs in Figure 5C). Of particular note in Figure 5C are the mRNAs encoding KTF1, NRPE1, DCL2, and AGO6, each of which function in RdDM (reviewed in Zhang et al., 2018). To determine whether RdDM-factor mRNAs were enriched among the 325 mRNAs identified in the ALY1-GFP RIP-seq, we performed a gene ontology (GO) term enrichment analysis. The ALY1-GFP–enriched GO categories centered on cytoskeleton and motor proteins, but not RdDM, small RNA production, or chromatin modification (Figure 5E), suggesting that ALY1 has no preference for mRNAs that generate our observed RdDM phenotype. Upon further examination of the same mRNA set, we identified several enriched 5- to 10-nucleotide RNA motifs (Supplemental Table 3). Six of these motifs had matches to binding sites of known RNA binding proteins (Ray et al., 2013), including two ALY/REF binding sites identified in human cells (Shi et al., 2017). We therefore hypothesize that these motifs represent putative ALY1 binding sites.

ALY1 Functions in mRNA Export

To identify any mRNAs with altered nuclear export in aly1 mutants, we performed mRNA-seq of cytoplasm- and nucleus-enriched fractions from two replicates of wild-type Col and aly1 (sequencing statistics in Supplemental Table 1). We confirmed the quality of our cytoplasmic and nuclear fractionation before (Supplemental Figure 6B) and after (Supplemental Figure 6C) sequencing, and this RNA was from the same extractions used for small RNA analysis shown in Figure 3E. In addition, Supplemental Figure 7 shows the high reproducibility between sequenced biological replicates. We analyzed the cytoplasm-enriched accumulation of mRNAs in aly1 compared with wild-type and identified mRNAs representing 3487 genes with more than twofold reduced cytoplasmic accumulation in aly1 (Figure 6A, red box; genes listed in Supplemental Data Set 2). We then tested the nuclear accumulation of these mRNAs (Figure 6B). We consider the simultaneously reduced nuclear and cytoplasmic mRNA levels to reflect overall reduced accumulation levels in aly1 cells. Those mRNAs with reduced overall accumulation included ALY1 itself, Repressor Of Silencing 1, and genes within the 12.6-kb deletion in the aly1-2 line (Figure 6B). Given the role of ALY1 in mRNA nuclear export, we were instead interested in the mRNAs that either had no change in accumulation in aly1 nuclei (2404), or showed increased nuclear accumulation (208) (Figure 6B; genes listed in Supplemental Data Set 2). Overall, we conclude that ALY1 functions in the nuclear export of 2612 mRNAs, which is significantly higher than the 325 mRNAs enriched by ALY1-GFP RIP-seq, suggesting that ALY1 affects the export of many mRNAs indirectly. By overlapping the mRNAs identified by RIP-seq in Figure 5C with the mRNA export data, we identified the ALY1-bound mRNAs that other ALY proteins could not compensate to export successfully in the aly1 mutant (Figure 6C). These included mRNAs that generate proteins with known roles in RdDM: AGO6, and DCL2.

Figure 6.

Analysis of mRNA Export in aly1 Mutants.

(A) mRNA-seq analysis of the cytoplasm-enriched fraction of wt Col and aly1 for all genes (no TEs). The number of genes represented in each section of the volcano plot is listed above. The red box indicates mRNAs with reduced cytoplasmic mRNA accumulation.

(B) mRNA-seq analysis of the nucleus-enriched fraction of wt Col and aly1 for the mRNAs with decreased cytoplasmic expression in aly1 from part A (red box).

(C) Scatterplot of wt Col and aly1 nuclear/cytoplasmic mRNA ratios of the mRNAs identified as enriched in the ALY1 RIP-seq.

ALY1 Enables Pol V Transcript Accumulation

We aimed to determine the mechanism of RdDM defect in aly1 mutants. First, since in human and mouse cells ALY/REF is known to interact with and export mRNAs through the RNA base modification of 5-methylcytosine (m⁵C; Yang et al., 2017), we determined whether m⁵C is necessary for the function of ALY1 in plants. We tested the known Arabidopsis m⁵C RNA methyltransferase TRM4b (David et al., 2017) for a role in RDR6-RdDM. TRM4b has thus far been implicated in mRNA export only in animal cells, and we did not detect a change in RDR6-RdDM in trm4b mutants (Supplemental Figure 3), demonstrating that RNA m⁵C does not function with ALY1 in RDR6-RdDM. Second, since ALY1 binds DCL2 mRNA (Figure 5B) and DCL2 mRNA fails to correctly export to the cytoplasm in aly1 mutants (Figure 6C), we tested whether DCL2 function is perturbed in aly1 mutants. However, we found that DCL2 functions correctly in aly1 mutants to generate 22-nucleotide siRNAs (Supplemental Figure 8).

Since aly1 mutants closely phenocopy the global loss of RdDM of pol V mutants, and since NRPE1 mRNA, encoding the Pol V largest subunit, is bound by ALY1 (Figure 5C), we assayed the accumulation of NRPE1 protein in aly1 mutants. We found that NRPE1 accumulated normally (Figure 7A); however, Pol V function was perturbed in aly1-2 mutants (Figures 7B and 7C). The failure of Pol V–generated scaffolding transcripts to accumulate from known target loci InterGenic Noncoding (IGN) Region 6 and IGN26 was not an indirect consequence of a failure of RdDM, as Pol V protein (Figure 7A) and transcripts (Figure 7B) continued to be generated in pol IV mutants that lost RdDM. Instead, aly1-2 resembled drd1 mutants, in which Pol V transcripts do not accumulate (Figure 7B; Wierzbicki et al., 2008). However, it is unlikely that the loss of RdDM in all aly1 mutants functions through a Pol V–related factor such as DRD1, as we tested Pol V function in the less severe aly1-1 mutant and found that failure of Pol V transcript accumulation was specific to the aly1-2 allele (Figure 7C). This result suggests that the defect in scaffolding transcript accumulation is not the root cause of RdDM failure in all aly1 mutants, and explains why the aly1-2 mutation has a more severe RdDM phenotype compared with aly1-1.

Figure 7.

ALY1 Is Required for Proper Pol V Function.

(A) Immunoblot of the largest subunit of the Pol V complex, NRPE1.

(B) and (C) RT-PCR of Pol V–generated transcript accumulation at IGN loci. ACTIN2 serves as a loading and positive control. The wt Col, aly1-2, and pol V samples displayed in (B) and (C) represent distinct biological replicates.

ALY1 Enables AGO6 Protein Accumulation

We next investigated AGO6, since the AGO6 mRNA failed to export from the nucleus in aly1-2 mutants (Figure 6) and was weakly (not statistically significant) enriched in the ALY1-GFP RIP-seq (Figure 5C). AGO6 is a key RdDM protein that functions in Pol IV-RdDM and RDR6-RdDM by incorporating 24- and 21- to 22-nucleotide siRNAs, respectively, and subsequently targeting a Pol V–generated scaffolding transcript to trigger DNA methylation (reviewed in Zhang et al., 2018). To determine whether the defect in AGO6 mRNA export results in reduced AGO6 protein accumulation, we performed an immunoblot for AGO6 whole-cell protein levels and found decreased AGO6 protein signal accumulation in six biological replicates of aly1 mutants compared with wild-type Col (Figure 8A). We confirmed the lack of normal AGO6 protein accumulation in both of the tested aly1 alleles (Figure 8B), and found that aly1 allele severity corresponded to AGO6 protein signal accumulation. The weaker aly1-1 allele retained 50 to 75% of the wild-type AGO6 levels, while the more severe aly1-2 allele retained only 25 to 50% (Figure 8B).

Figure 8.

AGO6 Protein Fails to Properly Accumulate in aly1 Mutants.

(A) Immunoblot of AGO6 and AGO4 protein levels in wild-type Col compared with aly1-2 mutants. The arrowhead points to AGO6, which is below a cross reactive nonspecific band. The top horseradish peroxidase immunoblot represents distinct biological replicates compared with the bottom fluorescence-detection immunoblot. AGO6 protein quantification is shown below, values are normalized to the ACTIN11 loading control. wt Col, wild-type Columbia.

(B) Quantification of AGO6 levels in aly1-1 and aly1-2 compared with percentages of the wild-type Col protein loaded. AGO6 protein quantification is shown below, values are normalized to the ACTIN11 loading control.

(C) Expression levels of the AGO6, AGO4, and ACTIN11 mRNAs mined from the mRNA-seq data in Figure 2A. Error bars represent the sd from the mean.

(D) Steady-state AGO6 mRNA accumulation level in lines carrying AGO6 transgenes measured by RT-qPCR. Error bars represent sd from the mean.

(E) Immunoblot of AGO6 protein levels in lines expressing AGO6 transgenes. The horseradish peroxidase immunoblot and fluorescence-detection immunoblot represent the same biological replicates. AGO6 quantification from the fluorescence-detection immunoblot is shown below, values are normalized to the ACTIN11 loading control.

AGO6 protein accumulation is reduced when siRNAs are absent (Havecker et al., 2010); however, this occurs on the protein level (with no change in AGO6 mRNA levels), whereas we detected altered accumulation of the AGO6 cytoplasmic mRNA (Figure 6C). This result suggests that the reduction in siRNA level in aly1 (Figure 3F) is not responsible for the lack of AGO6 protein accumulation. To test this idea, we monitored AGO4 accumulation in aly1 mutants, since the closely related AGO4 protein also fails to accumulate when siRNAs are absent (such as in pol IV mutants; Havecker et al., 2010). We did not detect a reduction in AGO4 levels in aly1 (Figure 8A), demonstrating that enough siRNAs are present in aly1 to support AGO4 and AGO6 protein accumulation. Therefore, the reduction of AGO6 protein is not due to deficient accumulation of siRNAs, but rather to the failure to efficiently export the AGO6 transcript itself.

To test whether the lack of AGO6 protein accumulation in the aly1 mutant could be rescued by overexpression of AGO6, we generated an AGO4 promoter-driven AGO6 coding region transgene (AGO4pro:AGO6), which has previously been shown to overexpress the AGO6 protein (Havecker et al., 2010). AGO4 has a much higher and more developmentally broad expression pattern compared with AGO6 (Figure 8C; Havecker et al., 2010; Eun et al., 2011). Our AGO4pro:AGO6 transgene was able to produce an overabundance of AGO6 mRNA and protein signal when ALY1 function is not perturbed (Figures 8D and 8E). However, when this same transgene was transformed into the aly1 mutant, the AGO6 transcript was overexpressed (Figure 8D), but there were not high levels of AGO6 protein (Figure 8E). We conclude that ALY1 functions to allow efficient AGO6 protein production from either the endogenous AGO6 gene, or from a distinct second copy driven by a strong promoter. In addition, since the AGO4pro:AGO6 transgene contains AGO4 5ʹ and 3ʹ untranslated region (UTR) sequences (see “Methods”), we conclude that ALY1 must function through a signal in the AGO6 exons or introns. Lastly, the AGO6 mRNA had several putative occurrences of the enriched RNA motifs identified via RIP-seq, most of which were in exons (Supplemental Table 3), correlating with the observed sufficiency of these regions for ALY1-dependent AGO6 production.

DISCUSSION

The ALY/REF family of animal and fungal proteins functions to export mRNAs from the nucleus to the cytoplasm for translation (Luo et al., 2001). In Arabidopsis, there are four ALY proteins that cooperatively perform this function (Pfaff et al., 2018). In support of their role in mRNA export, plant ALY2, ALY3, and ALY4 proteins interact with THO/TREX components (TEX1, UAP56, and MOS11) and are thus directly linked to known mRNA export machinery (Sørensen et al., 2017). In addition, Arabidopsis ALY1 complements the otherwise lethal yra1 mutation in yeast, indicating that its function in nuclear export is conserved (Pfaff et al., 2018). In this study, we find that ALY1 has a similar molecular phenotype as RdDM mutants. ALY1 mutants display a defect in de novo CHH methylation, TE transcriptional suppression, and the initiation of transgene silencing.

Of the three ALY1 alleles tested, aly1-2 displayed the most severe loss of RdDM (Figures 1A and 1C). We found that this was due the simultaneous loss of Pol V function (and therefore production of scaffolding transcripts) and the inefficient export of the AGO6 mRNA. Our finding that there are at least two methods of RdDM dysfunction in aly1-2 correlates with our repeated unsuccessful attempts to complement the aly1-2 mutant with a variety of ALY1 transgenes (data not shown). The less severe aly1-1 allele did not display a defect in Pol V function, demonstrating that this was not the root cause of RdDM failure in all aly1 mutants. Instead, both of the tested aly1 mutants had reduced levels of AGO6 protein, which is known to broadly function in RdDM and very specifically in the initiation of TE and transgene silencing (Zheng et al., 2007; Eun et al., 2011). The lack of AGO6 protein was likely due to the binding of the AGO6 mRNA by ALY1 (although this enrichment was not statistically significant in Figure 5C) and the subsequent inefficient nuclear export of the AGO6 mRNA (Figure 6C) for translation. The regulation of the AGO6 mRNA was observed independent of the promoter, UTRs, or locus from which the AGO6 transcript was generated; therefore, the regulation must be cued by signals in the exons or introns of the AGO6 transcript. In addition, it remains a possibility that ALY1 may enable RdDM through other mRNAs in addition to AGO6, representing a higher order deficiency where multiple factors required for RdDM are all simultaneously deficient.

Genetically, we find that mutation of ALY1 has a distinct consequence (loss of RdDM) that mutations of the other three ALY proteins do not; therefore, the ALY family of proteins must have at least partially nonoverlapping distinct mRNA targets. We theorize that the four Arabidopsis ALY proteins may contribute to mRNA export by acting on nonoverlapping sets of mRNAs to combine the full molecular function of the ALY/REF family. The sequence motifs, RNA localization, or other intrinsic features of individual RNAs that give rise to their different ALY protein specificity are currently enigmatic. Examination of the direct mRNA targets of ALY1 identified by RIP-seq demonstrated correlation to cytoskeleton and motor protein functions, and there are several enriched sequence motifs present, including two known binding sites of human ALY/REF. In addition, there is likely a mechanism of mRNA export from the nucleus that is distinct from ALY1 to ALY4, since even though the 4x aly mutant has a severe morphological phenotype, the plant remains viable and fertile (Pfaff et al., 2018).

Together, our data reconciles a very specific molecular phenotype (loss of RdDM) with a genetic mutation in a broadly acting mRNA export factor. Even though RdDM is a process that requires multiple types of noncoding RNAs (Wendte and Pikaard, 2017), and our reverse genetic screen could have identified any of these corresponding RNA binding proteins, instead we have identified a broadly acting RNA binding nuclear export factor. This provides an example of the difficulty of linking genotype and phenotype, as any one of the hundreds of mRNAs bound and exported by ALY1 could result in a phenotype, yet the phenotype observed can be largely attributed to the misregulation of one or two mRNAs.

METHODS

Plant Material

Arabidopsis (Arabidopsis thaliana) ecotype Col plants were grown under 18-h cool white (4100K) fluorescent light at 120 mmol m⁻² s⁻¹ provided by Sylvania Octron Eco bulbs at 22°C. All tissue is inflorescence (stages 1 to 12) immature flower buds unless otherwise specified. The aly1-1 allele is the SAIL_381_E08 insertion into ALY1 fourth exon. The aly1-2 allele is the SALK_034227 insertion into the ALY1 3′ UTR. We used CRISPR/Cas9 to generate the aly1-3 allele, which is homozygous for a small deletion just upstream of the RNA binding domain in the second exon of the ALY1 gene (details in Supplemental Methods). Other allele information is shown in Supplemental Table 4. The aly1-2 allele is used for all experiments unless otherwise specified.

Transgene Constructs

Transgenes and methodology used in Figure 4 were described previously (Fultz and Slotkin, 2017). The AGO4pro:AGO6 transgene was constructed by amplification of AGO6 coding sequence (including introns and without 5′/3′ UTRs) from Col genomic DNA. An N-terminal FLAG tag was added to AGO6 by inclusion in the forward primer. AGO4 UTR, promoter, and terminator sequences were used. The transgene was cloned using In-Fusion (Takara) into the vector pBGW. Cloning primers are shown in Supplemental Table 4.

DNA Methylation Analysis

Single-locus DNA methylation was assayed by bisulfite DNA conversion (Zymo Gold Kit, Zymo Research) followed by PCR amplification, cloning into pCR4 (Topo TA Cloning for Sequencing Kit, Life Technologies), and Sanger sequencing of at least six clones per PCR product/genotype. PCR primers are shown in Supplemental Table 4. The 95% Wilson confidence intervals were calculated as described previously (McCue et al., 2015).

MethylC-seq was performed as previously described (Panda et al., 2016). DNA was isolated from each sample using fractional precipitation followed by phenol-chloroform extraction and RNase A treatment. One microgram of isolated DNA was used to prepare libraries as previously described (Urich et al., 2015). Single-end Illumina sequencing of 150 bp was performed at the University of Georgia Genomics Facility using an Illumina NextSeq500 instrument. Data processing and downstream analysis are described in the Supplemental Methods. Genomic coordinates for the following regions are provided in Supplemental Data Set 3: Pol IV-RdDM TEs, RDR6-RdDM TEs, IR regions, met1-dependent TEs, cmt3-dependent TEs, cmt2-dependent TEs, cmt2 CHH hypo-DMRs, and pol V CHH hypo-DMRs.

Small RNA Analysis

RNA gel blotting was performed as in described previously (McCue et al., 2015). Four micrograms of small RNA was loaded per lane. The TAS3 and Athila6A probes were generated by randomly degrading a ³²P-labeled in vitro–transcribed RNA to ∼50-nucleotide fragments. PCR primers used to generate probe templates are listed in Supplemental Table 4. The SimpleHat2, U6, and tRNA^Met probes were generated by 5′ end labeling DNA oligonucleotides listed in Supplemental Table 4. Small RNA sequencing was performed as described previously (McCue et al., 2015). Library production, data processing, and downstream analysis are described in the Supplemental Methods.

RNA Sequencing

Total, nuclear, and cytoplasmic cell extracts were obtained as in the cellular fractionation protocol described previously (Park et al., 2005), with the following modification. Miracloth was substituted for using a 100-μm cell strainer. RNA was extracted using the TRIzol LS protocol (Invitrogen) and enriched for poly(A) mRNAs using the NEBNext Poly(A) mRNA Magnetic Isolation Module (NEB). Libraries were prepared using the NEBNext Ultra II Directional RNA library prep kit (NEB) and dual-indexed for Illumina sequencing on a HiSeq 4000 system. Data processing and downstream analysis are described in the Supplemental Methods.

RT-PCR

Total RNA was extracted using TRIzol reagent (Life Technologies). Ten micrograms of RNA was treated with TURBO DNase (Invitrogen), and then 1 μg of DNase-treated RNA was used as input into SuperScript III (Invitrogen) RT with an oligo(dT) primer. Quantitative PCR was performed as described previously (Nuthikattu et al., 2013).

For the RT-PCR in Figures 7B and 7C, RNA was extracted from 9-d-old seedlings, and the SuperScript IV One-Step RT-PCR Kit (Invitrogen) was used with gene-specific primers. Primers are listed in Supplemental Table 4.

RNA IP Sequencing

Two sets of three samples were prepared: each set included one wild-type Col mock IP sample and two biological replicates (different set of 20 plants in the same experiment) of ALY1-GFP IP. Set 1 was treated with RiboMinus Plant Kit for RNA-seq (Invitrogen) before library preparation, while set 2 samples did not undergo RiboMinus treatment. For each sample, 0.3 g of inflorescence tissue was used for protein extraction, and the IP was performed using 1 μg/IP αGFP (ab290, Abcam) at room temperature for 90 min (details in Supplemental Methods). The resulting RNA was either RiboMinus treated first or directly used as input into NEBNext Ultra II Directional Library Prep kit (NEB) and dual-indexed for paired-end Illumina sequencing on a HiSeq 2500 system. Data processing and downstream analysis are described in the Supplemental Methods.

Oligo(dT) Fluorescence In Situ Hybridization

Oligo(dT) in situ localization was performed as previously described (Parry et al., 2006), with the following modifications. Ten picomoles of 5ʹ Fluorescein-oligo-d(T)50 (GeneLink) was used, and after incubation, but before mounting, samples were washed once in 2× (SSC) + 0.1% SDS and once in 0.2× SSC + 0.1% SDS. Fluorescein was visualized using a 488 nm laser and 515/530 detector on a Nikon C1 confocal microscope using Nikon NIS-Elements and EZ-C1 software.

Immunoblot Analysis

Protein was extracted from tissue ground in liquid nitrogen using 20 mM Tris-HCl, pH 7.5, 5 mM MgCl₂, 300 mM NaCl, 0.1% Nonidet P-40, and 1% plant protease inhibitor (GoldBio). For immunoblotting, the following antibody concentrations in 3% milk and 1× PBS-Tween were used: 1:1000 αAGO6 (AS10-672, Agrisera), 1:10,000 αACTIN11 (AS10-702, Agrisera), 1:2000 αAGO4 (AS09-617, Agrisera), and 1:1000 αNRPE1. AGO6 immunoblot quantification was performed using LI-COR imaging software; AGO6 protein signal was normalized to loading control ACTIN11 for each respective lane, and the background signal detected in the ago6 lane was set as zero. Cellular fractionation of protein was performed as described by McCue et al. (2015))).

Chromatin IP

Chromatin IP experiments were performed as described previously (Huettel et al., 2006), with modifications described in the Supplemental Methods. Chromatin was immunoprecipitated using the antibody to H3K9me2 (Abcam) using 5 µg/IP, and immune complexes were collected using salmon sperm DNA-blocked protein A agarose beads (Millipore). Mock IPs (bead only) were performed for each sample and used for normalization of the quantitative (q)PCR. The qPCR was performed with primers shown in Supplemental Table 4. Results shown represent three independent biological replicates (independent pools of tissue) for each genotype. Data normalization was performed by calculating fold enrichment over a no-antibody mock IP.

Accession Numbers

Sequence data from this article can be found in the Arabidopsis Genome Initiative database under the following accession numbers: ALY1 (AT5G59950), AGO6 (AT2G32940), POL IV (NRPD1) (AT1G63020), POL V (NRPE1) (AT2G40030), RDR6 (AT3G49500), MET1 (AT5G49160), CMT3 (AT1G69770), CMT2 (AT4G19020), SGS3 (AT5G23570), SDE3 (AT1G05460), ALY2 (AT5G02530), ALY3 (AT1G66260), ALY4 (AT5G37720), HPR1 (AT5G09860), TEX1 (AT5G56130), TRM4B (AT2G22400), DDM1 (AT5G66750), DCL2 (AT3G03300), TAS3A (AT3G17185), SIMPLEHAT2 (AT5TE89325), and DRD1 (AT2G16390).

Genome-wide sequencing data sets used in this work are listed in Supplemental Table 1. Methylome, transcriptome, RIP-seq, and small RNA sequencing data sets can be accessed at the National Center for Biotechnology Information Gene Expression Omnibus as GSE118709.

Supplemental Data

• Supplemental Figure 1. Global loss of inverted repeat RdDM in aly1 mutants (supports Figure 1).

• Supplemental Figure 2. Loss of RdDM is specific to aly1 (supports Figure 1).

• Supplemental Figure 3. Loss of RdDM does not occur in other mRNA export mutants (supports Figure 1).

• Supplemental Figure 4. The deletion of six genes in the aly1-2 mutant line does not cause the loss of RdDM phenotype (supports Figure 2).

• Supplemental Figure 5. Flowering time analysis of aly1 mutant plants (supports Figure 3).

• Supplemental Figure 6. RIP-seq and nuclear/cytoplasmic RNA-seq controls and validation (supports Figures 5 and 6).

• Supplemental Figure 7. Correlation matrix showing expression profile similarity in nuclear and cytoplasmic RNA-seq libraries from aly1 and wt Col (supports Figure 6).

• Supplemental Figure 8. siRNA accumulation from all TAS regions in wt Col, aly1 and dcl2 (supports Figure 6).

• Supplemental Table 1. Deep sequencing statistics and availability.

• Supplemental Table 2. Genes with statistically significant elevated and reduced mRNA accumulation level in the aly1 mutant.

• Supplemental Table 3. Statistically significant motifs in the 325 statistically enriched mRNAs in the ALY1-GFP RIP-seq.

• Supplemental Table 4. Primer sequences and alleles used in this report.

• Supplemental Methods. Plant material, MethylC-seq data processing and analysis, small RNA analysis, RIP-seq, RNA-seq and RIP-seq data processing.

• Supplemental Data Set 1. Genes with statistically significant elevated and reduced mRNA accumulation level in the ALY1-GFP RIP-seq.

• Supplemental Data Set 2. Genes with reduced mRNA accumulation in aly1 cytoplasm, separated by their nuclear accumulation.

• Supplemental Data Set 3. Genomic coordinates of features.

Dive Curated Terms

The following phenotypic, genotypic, and functional terms are of significance to the work described in this paper:

• KTF1 Gramene: AT5G04290

• KTF1 Araport: AT5G04290

• CMT2 Gramene: AT4G19020

• CMT2 Araport: AT4G19020

• SDC Gramene: AT2G17690

• SDC Araport: AT2G17690

• RDR6 Gramene: AT3G49500

• RDR6 Araport: AT3G49500

• CMT3 Gramene: AT1G69770

• CMT3 Araport: AT1G69770

• DCL2 Gramene: AT3G03300

• DCL2 Araport: AT3G03300

• DRD1 Gramene: AT2G16390

• DRD1 Araport: AT2G16390

• MET1 Gramene: AT5G49160

• MET1 Araport: AT5G49160

• AGO4 Gramene: AT2G27040

• AGO4 Araport: AT2G27040

• TAS3 Gramene: AT3G17185

• TAS3 Araport: AT3G17185

• AGO6 Gramene: AT2G32940

• AGO6 Araport: AT2G32940