RNA 3′ processing functions of Arabidopsis FCA and FPA limit intergenic transcription

Abstract

The RNA-binding proteins FCA and FPA were identified based on their repression of the flowering time regulator FLC but have since been shown to have widespread roles in the Arabidopsis thaliana genome. Here, we use whole-genome tiling arrays to show that a wide spectrum of genes and transposable elements are misexpressed in the fca-9 fpa-7 (fcafpa) double mutant at two stages of seedling development. There was a significant bias for misregulated genomic segments mapping to the 3′ region of genes. In addition, the double mutant misexpressed a large number of previously unannotated genomic segments corresponding to intergenic regions. We characterized a subset of these misexpressed unannotated segments and established that they resulted from extensive transcriptional read-through, use of downstream polyadenylation sites, and alternative splicing. In some cases, the transcriptional read-through significantly reduced expression of the associated genes. FCA/FPA-dependent changes in DNA methylation were found at several loci, supporting previous associations of FCA/FPA function with chromatin modifications. Our data suggest that FCA and FPA play important roles in the A. thaliana genome in RNA 3′ processing and transcription termination, thus limiting intergenic transcription.

The Arabidopsis thaliana proteins FCA and FPA were identified through their effects as regulators of the floral transition (1). They were classified as functioning in the autonomous floral pathway, which comprises a set of activities that promotes flowering by down-regulating expression of the MADS box floral repressor FLOWERING LOCUS C (FLC) (2). FCA and FPA are RNA recognition motif (RRM)-containing proteins (3, 4), which negatively regulate their own expression through promotion of an internal polyadenylation site (5, 6). FCA, but not FPA, physically interacts with FY, a protein homologous to the RNA 3′ processing component named Pfs2p in Saccharomyces cerevisiae and WDR33a in humans (7, 8). An extensive suppressor mutagenesis analysis combined with detailed molecular analysis revealed how FCA and FPA down-regulate expression of FLC (9). Both were found to promote proximal polyadenylation of an FLC antisense RNA (6, 9), and this leads to down-regulation of FLC sense transcription in a mechanism involving the activity of the histone 3-lysine 4-demethylase (FLD) (9). The mechanism linking alternative polyadenylation of the antisense transcript, histone demethylation of the locus, and down-regulation of FLC sense transcription remains to be resolved.

A more general genomic role for FCA and FPA was suggested when they were identified in a genetic screen targeting factors required for transgene-induced silencing. fca and fpa mutants suppressed systemic silencing and DNA methylation of the homologous endogenous A. thaliana gene in response to a mobile RNA silencing signal generated from a hairpin construct (10). FCA and FPA were also shown to be required for DNA methylation changes at low copy transposon and retrotransposon sequences in the A. thaliana genome (10), although a general role for the autonomous pathway in the repression of gene expression through DNA methylation does not seem to be the case (11). RNA 3′ processing and chromatin modification have been linked previously in a study of the yeast Paf chromatin complex (12). Mutations in core 3′ processing components (CstF64, symplekin, and CPSF100) are also thought to trigger chromatin silencing through generation of aberrant RNA substrates (13). However, to further investigate this link, we undertook a genome-wide analysis to identify the extent of FCA and FPA regulation in A. thaliana. Here, we describe our analysis using whole-genome tiling arrays of A. thaliana, where we characterize the misexpression profile in an fca-9 fpa-7 (fcafpa) double mutant. A number of genes and transposable elements were misexpressed, with a significant bias for genomic segments mapping to the 3′ region of genes. In addition, a large number of previously unannotated (UA) genomic segments corresponding to intergenic regions were misexpressed in the double mutant. Characterization of these UA segments suggests that FCA and FPA play important genome-wide roles in 3′ processing and transcription termination, and in several cases, their function overlaps with other pathways mediating chromatin regulation.

Results

Wide Range of Protein-Coding Genes, Transposable Elements, and UA Segments Are Misexpressed in fcafpa.

To investigate the extent of FCA and FPA targets in the A. thaliana genome, we used the Affymetrix Whole Genome Tiling Array Chip AtTile1R [Gene Expression Omnibus (GEO) accession no. GPL1980]. We compared genome-wide transcript levels in three biological replicates of WT Columbia (Col-0) and fca-9 fpa-7 (fcafpa) double mutant seedlings. Two developmental stages, 7 and 17 d after germination, were analyzed. The AtTile1R array has 6,184,720 perfect match (PM) and mismatch (MM) 25-mer probes and was originally designed to tile the Institute of Genomic Research (TIGR) version 5 of the A. thaliana genome. Initial selections of differentially expressed probes were made with cutoffs of Benjamini–Hochberg-adjusted P value < 0.05 (14) and (absolute) fold change > 1.5.

Overall, 848,324 (∼14%) probes were differentially expressed in 17-d fcafpa seedlings, and of these, ∼8% were up-regulated and 6% were down-regulated. In contrast, only ∼9% of the probes (4.7% up-regulated and 4.3% down-regulated) were misregulated in 7-d fcafpa seedlings. These results suggest a developmental bias in expression profiles of fcafpa mutants. We remapped all PM and MM probes against TAIR Genome version 9 (TAIR9) (15) using the short-oligonucleotide alignment program (SOAP) (16). From this dataset, we computed sets of segments that represent contiguous series of probes with respect to their genomic positions, which were significantly misexpressed (up-regulated or down-regulated in fcafpa) compared with WT Columbia. After alignments of the segments to the respective annotations, as defined in the Genomic File Format (GFF) annotation files from TAIR9, the up-regulated and down-regulated segments in fcafpa fell into different annotation groups (Table 1); 2.4% and 2% of the estimated 25,000 genes of the A. thaliana genome were up-regulated and down-regulated, respectively, in 7-d fcafpa seedlings, and these numbers increased significantly in 17-d fcafpa seedlings (∼5% and 10%, respectively). Comparison of the two developmental stages revealed that 139 genes were commonly up-regulated and 320 were down-regulated (23% and 63% of those listed at 7 d, respectively).

Table 1.

Summary of the differentially expressed segments in the tiling array

	Day 7		Day 17
TAIR9 annotation	Up	Down	Up	Down
Protein-coding gene	598	505	1,139	2,813
Pseudogene	5	0	11	0
Transposable elements	43	20	102	59
Noncoding RNA	6	13	7	21
tRNA	0	0	1	0
miRNA	1	0	0	1
snoRNA	0	0	0	3
Unannotated	82	50	95	54

There was a bias in misregulation of segments corresponding to 3′ ends of genes compared with 5′ ends. We compared the TAIR9 GFF annotation files for 3′ and 5′ UTRs of genes with gene model annotations separately. At 7 d, the total number of misregulated genes in the 5′ and 3′ UTRs varied greatly (∼7% and 40%, respectively). Examples of misregulated 3′ UTRs are genes At1g01020 (ARV1) and At3g23100 (XRCC4), both of which have two alternative transcripts in TAIR9 (http://www.arabidopsis.org) (Fig. S1). Among nonprotein-coding genes, the MIR398C microRNA (miRNA) precursor was up-regulated at 7 d, whereas that for MIR863A was down-regulated at 17 d, and three small nucleolar RNAs (snoRNAs) were up-regulated at 17 d. The most unexpected class in fcafpa was the relatively large number of previously unannotated segments of the genome, and we focused our subsequent analysis on these, including exploring potential links to other gene silencing pathways. Of the 82 UA segments, we attempted to verify 27 and confirmed 17 (∼65%). In contrast, of the 13 transposable elements (TEs) tested, only 4 were confirmed (∼31%). Closer analysis of the microarray plots suggested that the false positives among TEs were because of the extensive cross-hybridization of the probes with other repetitive regions.

Majority of UA Segments Map to 3′ Ends of Annotated Genes or Pseudogenes.

The tiling array had been hybridized with oligo d(T)-primed cDNA and converted to dsDNA by a DNA polymerase (17). It, therefore, preferentially detects polyadenylated transcripts and does not distinguish between strands. A number of reports emphasize the importance of antisense noncoding RNA in gene regulation in yeast, mammals, and plants (18–22), and FCA and FPA are both involved in FLC antisense transcript processing in A. thaliana (9). Therefore, we investigated the strand that was misregulated in fcafpa. Of the 17 PCR-verified UA segments, 14 were primarily up-regulated in one strand, whereas 3 were up-regulated on both strands. The corresponding genomic coordinates of the segments are indicated in the 5′ to 3′ direction of the transcripts in Table S1.

To assess the individual quantitative contributions of FCA and FPA, we compared expression of 15 of the up-regulated UA segments in the double fcafpa with single fca-9 and fpa-7 mutants (Fig. 1). The relative fold change in the double mutant ranged from 7 to 600 (Fig. 1A), whereas in fpa-7, it ranged from 1.5 to 200 and in fca-9, from 0.2 to only 3.5. fpa-7 had a much bigger quantitative effect on the UA segment expression than fca-9, and comparison of the effects of the two single mutants showed different quantitative effects of each at different loci (Fig. 1 B and C). The simplest explanation for these observations is that the targets that we have focused on are predominantly FPA-dependent. However, the situation seems more complex, because misexpression of some of the targets was much higher in the fcafpa double compared with either single mutant, and in at least two cases, down-regulation was observed in the fca single mutants, which indicates an antagonistic effect between FCA and FPA activity. Therefore, it seems likely that there is a complex relationship between FCA and FPA, which can act both independently, cooperatively, or antagonistically, depending on the target.

Fig. 1.

qRT-PCR analysis of UA segments in (A) fca-9 fpa-7, (B) fpa-7, and (C) fca-9 mutants. Histograms show mean values ± SEM for three independent PCR amplifications on at least three biological replicates. The y axis shows the fold change in C and fold change on a logarithmic (base 10) scale in A and B relative to WT Columbia (Col = 1) after normalization to UBC gene expression.

Analysis of the UA segments using the TAIR GBrowse tool revealed that most mapped to within 500 bp of the 3′ end of the closest annotated gene, with 14 segments within 200 bp. This paralleled the observed bias for misregulation at the 3′ end of genes. To understand this better, we analyzed the 15 verified up-regulated segments either using RT-PCR followed by sequencing (Fig. S2) or 5′ and 3′ rapid amplification of cDNA ends (RACE) analysis (Fig. S3). This revealed a complex picture, with every misexpressed segment resulting from a slightly different event at each locus. Representative examples from the complete set (Figs. S2 and S3) are shown in Fig. 2, with the gene annotation and altered expression shown for the different segments. Loss of FCA and FPA led to increased transcript read-through that continued several kilobases downstream to the adjacent gene (UA2 and UA10B) (Fig. 2 A and D), through an adjacent gene (UA4) (Fig. 2B), into an intergenic sequence (UA5) (Fig. 2C), into transposon-rich regions (UA15) (Fig. 2E), or in convergently transcribed genes (UA23 and UA24) (Fig. 2F). Northern analysis of the misexpressed transcripts suggested that loss of FCA and FPA resulted mainly in increased transcriptional read-through rather than a binary switch in use of polyadenylation sites, because the shorter transcript was still the predominant form in fcafpa (Fig. S4). Northern blot also revealed accumulation of high molecular-weight transcripts in fpa single and fcafpa double mutants, which was not detected by the RT-PCR or 3′ RACE analyses.

Fig. 2.

Schematic representation of a selection of verified UA segments and their locations in the A. thaliana genome (A–F). The schemes are adapted from http://neomorph.salk.edu/epigenome/epigenome.html with minor alterations. The blue arrows designate the UA segments, with the arrowheads pointing to the 3′ end of the transcript, green boxes are exons, red boxes are 5′ and 3′ UTRs, thin gray lines are introns, and gray boxes are nongenic annotations (i.e., transposons and pseudogenes). A′–F′ are plotted based on sequencing either the RT-PCR products (B′, C′, E′, and F′) amplified with primer pairs indicated in Fig. S2 or 5′ and 3′ RACE products (A′ and D′). The dashed lines and asterisk in A′ indicate the ∼3-kb intron splicing event; two asterisks in F′ designate the predicted outcome of loss of fcafpa on the affected genes. The TAIR9 annotation numbers are designated above or below each annotation. Different scales are used for the different gene representations.

Transcriptional Read-Through Is Frequently Associated with Large Intron Splicing.

In the majority of cases, the UA segments correspond to an alternatively polyadenylated transcript (e.g., UA2 corresponds to an alternatively processed form of At1g28140) with only 42 bp of the original 3′ UTR retained and the new polyadenylated transcript acquiring a 300-bp alternative 3′ UTR that terminates 54 bp away from the 3′ end of the adjacent gene, At1g28135 (Fig. 2A). UA4 represents a read-through transcript of At1g55805, which encompasses a differentially spliced version (compared with the TAIR9 annotation) of the downstream gene At1g55800 (Fig. 2B). UA5 and UA10B correspond to polyadenylated read-through transcripts of At1g62820 and At2g23780, respectively (Fig. 2 C and D), both extending ∼1 kb downstream. The alternative transcript of UA10B terminates very close (183 bp away) to the 5′ end of the next gene and a transposon mapping on the other strand. UA15 corresponds to read-through of At4g04640, a gene embedded in a transposon-rich genomic region, and UA23 and UA24 are overlapping read-through products of two convergently transcribed genes on chromosome 4 (Fig. 2 E and F). Many other examples are shown in Figs. S2 and S3.

In many cases, the extensive transcriptional read-through is associated with alternative splicing. In some situations, a very large intron (∼3 kb) is spliced out in the alternative transcript (UA2, UA3, and UA16) (Fig. 2A and Fig. S2 A and E). These represent some of the largest introns known in the A. thaliana genome. Interestingly, in UA3 and UA16, the protein-coding sequence is also changed because of the splice site choice upstream of the translation stop codon. In other cases, multiple splice variants of the alternatively polyadenylated transcript are observed (UA21) (Fig. S2G). In summary, transcriptional read-through, alternative polyadenylation, and alternative splicing result in apparently unannotated genomic segments being misexpressed in the fcafpa double mutant.

At first glance, not all UA segments seemed to be easily explained by transcriptional read-through from an upstream gene (for example, UA3 and UA228) (Figs. S2 and S3). UA3 is an antisense transcript extending 92 bp into the 3′ UTR of At1g52650 that is confirmed by 3′ RACE. (Fig. S2A). The gene upstream of UA3 that seemed most likely to generate a read-through transcript is transcribed in the other orientation. However, a more detailed analysis of UA3 transcripts shows that it is a read-through product of At1g52670, ∼2.3 kb away, with splicing of a large intron removing the sequence encoding the entire At1g52670 gene (Fig. S2A′). In this respect, UA3 is reminiscent of the XRCC4 example described earlier (Fig. S1B). Another example is UA228, which was found by 5′ and 3′ RACE analysis to correspond to an antisense transcript that spanned an annotated insertion and part of a helitron sequence (Fig. S3D). 3′ RACE analysis revealed alternatively spliced transcripts that were polyadenylated at multiple sites, but the poly (A) site used did not seem to differ between Col-0 and fcafpa. It is currently unclear if the primary basis of UA228 misexpression in fcafpa is because of transcriptional read-through.

UA Segments Can Be Direct FCA/FPA Targets and Are the Result of a Complex Interplay Between FCA and FPA Functions.

Because loss of FCA and FPA led to increased transcriptional read-through, we asked if this was a result of a direct association of FCA and FPA with the 3′ region of the gene. ChIP analysis showed that At1g28140 and At2g23780 are direct targets of FCA and FPA (Fig. 3). FCA interacts with At1g28140 and At2g23780 around the proximal poly (A) sites (Fig. 3B), whereas FPA interacts with the proximal poly (A) site of At2g23780, but its interaction with At1g28140 seems to be downstream of the proximal poly (A) site (Fig. 3C). These distinct positions were also found at FLC and FPA (6, 23), reinforcing the view that FCA and FPA participate in different mechanisms to promote 3′ processing and transcription termination.

Fig. 3.

ChIP analysis of genomic regions around At1g28140 and At2g23780. (A) A schematic of the genomic regions, (B) enrichment of FCA in Col-0 compared with fca-9, and (C) FPA::YFP compared with YFP after immunoprecipitation with an FCA or GFP antibody, respectively, are shown. Histograms show mean values ± SEM for enrichment calculated by percent input normalized against actin for three qPCR amplifications and two biological replicates. qPCR primers for UA2, At1g28140, UA10B, and At2g23780 are designated by solid bars on the gene models in A. Table S3 lists the primer sequences.

The transcriptional read-through and use of the downstream polyadenylation sites significantly reduced expression of associated genes (e.g., At1g28140 associated with UA2 and At2g23780 associated with UA10B) (Fig. S4D). Down-regulation of the genes, particularly for At1g28140, was stronger in the fcafpa double mutant than in either single mutant, suggesting a partially redundant role for FCA and FPA in this regulation. The relative changes in expression of the genes did not match the changes in UA expression, suggesting a complex interplay between transcription and RNA turnover pathways. Several mechanisms might link the length, sequence context, and secondary structure of 3′ UTR with reduction in overall expression of the whole transcript, including increased nonsense-mediated decay (24), inclusion of an miRNA site (25, 26), or RNA silencing mechanisms recognizing aberrant nascent transcripts (27). For At1g28140, it might involve a transcriptional silencing mechanism, because the TAIR GBrowse shows that the adjacent gene At1g28135 is associated with siRNAs and DNA methylation. RT-PCR analysis with primers situated inside At1g28135 and UA2 detected a PCR product in Col-0 cDNA, suggesting that low-level read-through already occurs in WT plants.

Overlap of FCA/FPA Activity with Other Silencing Pathways.

We had previously shown that FCA and FPA are involved in the regulation of expression of low-copy transposon and retrotransposon sequences in the A. thaliana genome and that this regulation is associated with changes in DNA methylation in a locus-specific manner (10). One of the UA segments, UA228, mapped within a helitron sequence (At1TE93275); therefore, we addressed whether FCA and FPA play a role in the complex interplay of silencing at this locus. Indeed, we found that UA228 was ∼100-fold up-regulated in a selection of RNA silencing mutants including ddm1, nrpd1a, and mom1 (Fig. 4). UA228 expression was exceedingly variable, affected by growth conditions and developmental stage, and differentially influenced by single or double loss or gain of FCA and FPA function. How these different pathways interact to regulate the common targets will require a much more extensive double mutant analysis. However, we showed by McrBC digestion assay that DNA methylation was significantly reduced in fcafpa at UA228 (Fig. 5A). Bisulfite sequencing revealed that FCA/FPA-dependent loss of DNA methylation at this locus was in all three sequence contexts (i.e., CG, CNG, and CHH) (Fig. 5B and Fig. S5A).

Fig. 4.

qRT-PCR analysis of the segment UA228 in various RNA silencing mutants. Histograms show mean values ± SEM for three independent PCR amplifications on two biological replicates. The y axis shows the relative fold change in logarithmic (base 10) scale normalized to UBC gene expression. Table S3 lists the mutant alleles.

Fig. 5.

DNA methylation analysis at the UA228 segment in Col vs. fcafpa by (A) McrBC digestion. Equal amounts of McrBC-digested (+) and undigested (−) DNA from two biological replicates of Col and fcafpa were amplified by PCR. (B) Bisulfite sequencing.

This result led us to investigate the WT DNA methylation status of the rest of the UA segments (http://neomorph.salk.edu/epigenome/epigenome.html). A high proportion of the genomic regions represented in the UA segments or their associated genes were shown to have DNA methylation or siRNA production (28) (Table S1). To explore the potential effect of loss of FCA and FPA in DNA methylation in conjunction with read-through transcription, we analyzed DNA methylation in part of the read-through transcript of UA4 by bisulfite sequencing. We picked this region also because of low genomic complexity (i.e., no transposons or pseudogenes in the close vicinity) and an interesting splicing pattern. Our data showed a significant increase in DNA methylation at two consecutive CG sites within At1g55800 in fcafpa, a twofold increase in CHH, and no change in CNG methylation (Fig. S5B).

In the light of our recent understanding of FCA and FPA function, we reanalyzed the DNA methylation in AtSN1 based on our previous data and found that, similarly, there was a significant increase in CHH methylation at a certain region within AtSN1 in fcafpa (Fig. S5C). Interestingly, we found that DNA methylation in another region in the AtSN1 locus was significantly higher in fca but not fpa and only slightly higher in fcafpa (Fig. S5C). AtSN1 has been shown to be up-regulated in fpa because of read-through transcription from an upstream gene (6). Overall, we conclude that FCA and FPA alter DNA methylation differentially at different loci. A more detailed analysis will reveal the interaction of FCA/FPA with DNA methylation in 3′ processing and transcription.

FCA and FPA activity on FLC is mediated through a histone demethylase, FLD (23). We tested the generality of the overlap of misexpression in fld and fcafpa by quantitative RT-PCR (qRT-PCR). Two of the UA segments analyzed, UA228 and UA234, were strongly up-regulated (∼150- and ∼10-fold, respectively) in fld-4. At1g62820, associated with UA5, also showed a twofold up-regulation in fld-4 (Fig. 4). Taken together, our data suggest that extensive read-through transcription because of defective 3′ processing is associated with chromatin changes and that FCA and FPA are involved in the interplay of these cotranscriptional mechanisms.

Discussion

The Arabidopsis RNA-binding proteins FCA and FPA were initially identified because of their role in flowering time, but subsequently, fca and fpa mutants were found to release silencing of a number of transgenes, transposons, and retroelements in the A. thaliana genome (10). FCA and FPA were, therefore, proposed to perhaps recognize some RNA feature and interact in a locus-dependent manner with RNA-mediated DNA methylation pathways. We have pursued the wider role of FCA and FPA in the A. thaliana genome by analyzing genome-wide transcriptional changes using the Affymetrix Whole Genome Tiling Array. Loss of FCA and FPA led to misexpression of a range of genes and noncoding sequences, with a high number of misexpressed unannotated genomic segments. FCA and FPA have been shown to regulate choice of poly (A) sites in their own transcripts (5, 6), and therefore, it would have been interesting to ask whether all or only a fraction of the exons were misexpressed. However, the signal from the arrays was too noisy to conclude the extent of FCA- and FPA-mediated alternative polyadenylation within genes without extensive verification. We, therefore, chose instead to characterize a subset of the UA segments in detail. Our conclusion was that these were, in general, connected to read-through transcripts of adjacent genes, with FCA and FPA promoting use of proximal polyadenylation sites. Loss of their activities resulted in read-through transcription associated with use of a downstream polyadenylation site and in many cases, alternative splicing as well. This changed termination profile was often associated with reduced expression of the corresponding gene and changed chromatin modifications. FCA and FPA, therefore, seem to play important genome-wide roles in cotranscriptional RNA processing and suppression of intergenic transcription. In several genomic regions, the read-through transcription was linked to chromatin modification.

The increased transcriptional read-through in fcafpa mutants suggests that FCA and FPA might promote use of otherwise weak poly (A) sites, perhaps through cis element recognition and/or enhanced trans factor binding. The downstream polyadenylation site used in the absence of FCA and FPA is, in many cases, several kilobases downstream (e.g., At1g28140, At1g55805, At1g52670, At4g24660, and XRCC4), and therefore, in the absence of FCA and FPA, considerable intergenic transcription occurs. Lack of 3′ cleavage and polyadenylation prevents termination of polymerase activity. In the torpedo termination model, a 5′ exonuclease recognizes the free 5′ end after cleavage and polyadenylation specificity factor (CPSF)-induced cleavage, chases, and then, displaces the polymerase (29). FCA and FPA associated with different regions at the 3′ ends of the genes and the individual fca and fpa mutants led to differential accumulation of the UA segments, despite similar quantitative effects on expression of the corresponding gene, suggesting that FCA and FPA influence the termination process in very different ways. The very high accumulation of the UA segments in fpa perhaps suggests increased read-through in fpa compared with fca or reduced turnover of transcripts in fpa that are normally rapidly degraded in WT or fca.

The extended read-through transcripts are, in many cases, associated with large introns (∼3 kb), some of the biggest found in the A. thaliana genome. The 3′ acceptor sites are generally in close proximity to the poly (A) site, suggesting that there may be a close connection between the choice of the 3′ acceptor site and the choice of poly (A) site. The fact that these FCA/FPA targets are often associated with chromatin modifications suggests a tight coupling between cotranscriptional RNA metabolism, chromatin modification machinery, and perhaps, transcription. This connection has previously been revealed by the A. thaliana mom1 mutant, which derepresses expression in the absence of DNA methylation changes (30) and is required for RNA turnover when 3′-end sequences are missing from plant transgenes (31). The yeast Paf1 elongation complex also affects 3′-end formation and chromatin modification (32), and therefore, it will be interesting to investigate how FCA/FPA function intersects with Paf function on plant genes. Transcriptional read-through in Paf mutants triggers nonsense-mediated decay pathways in yeast (24), and this would be consistent with the reduced levels of genes corresponding to the UA segments (At1g28140, At1g55805, At2g16510, At2g23780, At4g24660, and ARV1). Whether this type of mechanism is developmentally regulated to provide genome-wide coordination of gene expression changes remains to be established. As judged by FCA autoregulation, there is a reduction in FCA activity at a certain stage of development (33), and this could potentially lead to global transcription/termination/polyadenylation changes similar to those observed during mouse embryonic development (34).

In conclusion, this study has revealed the complexity in RNA 3′-end processing and transcription termination in the A. thaliana genome. The RNA-binding proteins FCA and FPA promote use of proximal polyadenylation and 3′ processing sites, and in their absence, intergenic transcription increases. Read-through transcripts are polyadenylated at alternative downstream sites that do not require FCA or FPA function, raising the question of what determines FCA and FPA specificity. The connections between transcription, cotranscriptional processing, and RNA quality control mechanisms are emerging from studies in many organisms and seem conserved. Continued analysis of the FCA and FPA mechanism of action should contribute to this area.

Materials and Methods

Plant Material and Growth Conditions.

Plants were grown in long-day conditions (16 h light and 8 h dark) in compost mix at 23 °C or on GM minus glucose plates at 20 °C and harvested 7 d after germination for subsequent RNA analysis. All mutants used in this study are in Columbia background. The list of mutants used in this study can be found in Table S2. The constructs overexpressing the full-length FCA transcript and the FPA:YFP transgene were described in refs. 3 and 10, respectively.

Tiling Array Analysis.

The tiling array was performed as described in refs. 17 and 35. SI Materials and Methods has a detailed explanation of the statistical analysis of the tiling array data.

RNA Extraction and Analysis.

SI Materials and Methods has information on RT-PCR, 5′ and 3′ RACE, and Northern blot analyses. Primers used in this study are listed in Table S3.

ChIP Analysis.

The ChIP was performed as described in ref. 23 using anti-FCA antibody or anti-GFP antibody (Invitrogen) for FCA and FPA–YFP analysis, respectively. The PCR amplification was carried out quantitatively using the Roche Lightcycler 480 Real Time Instrument. Relative enrichment was calculated using the percent input method followed by normalization against actin. ChIP primers are listed in Table S3.

DNA Methylation Analysis.

McrBC (New England Biolabs) digestion was performed according to the manufacturer's directions. For bisulfite sequencing, DNA was bisulfite-converted with the Epitech Bisulfite conversion kit (Qiagen). PCR primers used for amplification are listed in Table S3. PCR products were TOPO-TA–cloned (Invitrogen), and at least 22 clones in total for each genotype were sequenced from two biological replicates. Sequencing results were analyzed by KisMeth Software (36).