RNA Polymerase IV Functions in Paramutation in Zea mays

See allHide authors and affiliations

Science  27 Feb 2009:
Vol. 323, Issue 5918, pp. 1201-1205
DOI: 10.1126/science.1164508


Plants have distinct RNA polymerase complexes (Pol IV and Pol V) with largely unknown roles in maintaining small RNA–associated gene silencing. Curiously, the eudicot Arabidopsis thaliana is not affected when either function is lost. By use of mutation selection and positional cloning, we showed that the largest subunit of the presumed maize Pol IV is involved in paramutation, an inherited epigenetic change facilitated by an interaction between two alleles, as well as normal maize development. Bioinformatics analyses and nuclear run-on transcription assays indicate that Pol IV does not engage in the efficient RNA synthesis typical of the three major eukaryotic DNA-dependent RNA polymerases. These results indicate that Pol IV employs abnormal RNA polymerase activities to achieve genome-wide silencing and that its absence affects both maize development and heritable epigenetic changes.

In maize, mouse, and other eukaryotes, paramutation refers to a process by which heritable changes in gene regulation are facilitated by interactions between alleles on homologous chromosomes (1). As typically described, alleles conferring relatively high gene action invariably change to a repressed expression state when heterozygous with specific alleles or allelic states (1). Operationally, paramutation violates the first law of Mendelian inheritance that alleles segregate unchanged from a heterozygote and thus has important implications for normal genome function and evolution (2, 3), though few examples have proved experimentally tractable.

In maize, paramutations occurring at the Pl1-Rhoades (Pl1-Rh) allele of the purple plant 1 (pl1) locus involve at least four genes (46), two of which appear to be part of a small interfering RNA (siRNA) heterochromatin pathway (79). Repressed expression states of Pl1-Rh resulting from paramutation are heritably maintained by Required to Maintain Repression1 (RMR1), a previously undiscovered Sucrose Nonfermenting 2 (SNF2)–like adenosine triphosphatase, and Mediator of Paramutation1 (MOP1), a putative RNA-dependent RNA polymerase (RDR) (4, 8, 9). RMR1 and MOP1 appear to be similar to proteins involved in an RNA-directed DNA methylation (RdDM) pathway in the eudicot Arabidopsis thaliana (7). RMR1 is most similar to Arabidopsis CLASSY1 (CLSY1) and DEFECTIVE IN RNA-DIRECTED METHYLATION1 (DRD1) (5, 7), whereas MOP1 appears orthologous with Arabidopsis RDR2 (8, 9). In Arabidopsis, RDR2- and DICER-LIKE3-dependent 24-nucleotide (nt) siRNAs associate with a specific argonaute protein (AGO4) and are thought to guide, via sequence homology, the deposition of de novo cytosine methylation marks associated with transcriptionally repressed heterochromatin (10). The majority of genomic loci targeted by RdDM, and related small RNA (sRNA)–based silencing pathways in other eukaryotes, are transposable elements and other repetitive sequences (10, 11), which suggests an interplay between paramutation-induced repression and the means by which eukaryotic genomes cope with potentially deleterious repetitive features.

A terminal fragment of a CACTA-like DNA transposon related to the doppia subfamily (12) resides immediately 5′ of the Pl1-Rh coding region and is targeted by an RdDM-type pathway (7). Plants homozygous for rmr1, mop1, or rmr6 mutations have hypomethylated cytosines across this doppia fragment relative to heterozygous siblings (7), and RMR1 is also required for accumulation of doppia-like small (∼24 nt) RNA species (7). However, the functional connection between doppia cytosine methylation and Pl1-Rh paramutation remains unclear because no correlations between cytosine methylation levels and Pl1-Rh expression levels in nonmutant plants have been observed (7). In contrast, rmr6 mutant plants have increases in Pl1-Rh transcription rates that are correlated with 5′ hypomethylation in all cytosine contexts within, 5′ of, and immediately 3′ of the doppia fragment (6, 7).

Many of the Arabidopsis sRNA-associated repression mechanisms, including RdDM, are hypothesized to be dependent on the actions of two presumed DNA-dependent RNA polymerase complexes, Pol IV (Pol IVa) and Pol V (Pol IVb) (10, 13). All plants appear to have a conserved Pol IV largest subunit (RPD1) derived from an ancient duplication of the Pol II largest subunit (RPB1), and all flowering plants appear to have an RPD1 duplicate called RPE1 (14). The Arabidopsis RPD1 (AtNRPD1a) and RPE1 (AtNRPD1b) form two functionally distinct multisubunit complexes (Pol IV and Pol V, respectively) that share a single subunit, AtNRPD2 (13). Although both polymerase forms appear to be related to Pol II, neither the ability to produce an RNA transcript nor the exact nature of any potential nucleic acid template is known for the Pol IV complex (13), whereas recent evidence does indicate the involvement of Pol V in generating low-abundance RNAs (15). Additionally, knockout mutants show that both Pol IV and Pol V are nonessential for normal plant development in Arabidopsis (13).

Because rmr6 mutants are defective for both paramutation and proper maize development (6, 16) (fig. S1), we set out to identify the gene product. We mapped rmr6 to a ∼450-kb interval on maize chromosome 1L, which is syntenic with a region on rice chromosome 8 (fig. S2A) and contains six contiguous maize-specific gene models that are absent from the syntenic rice interval (fig. S2B) (17). Four of these models coalesced to a single, putative, 18-exon gene of ∼65 kb (Fig. 1A) with the potential to encode a 1444–amino acid polypeptide with similarity to NRPD1a, the largest subunit of Arabidopsis Pol IV (Fig. 1B) (18). Because Pol IV acts in an RdDM-type pathway to maintain epigenetic repression of genomic targets (13), DNA from this candidate gene was amplified and sequenced in homozygous rmr6 mutant plants of four different ethyl metanesulfonate–derived rmr6 alleles (17). These efforts identified single transition–type lesions within the predicted coding regions of the NRPD1a-like candidate gene (Fig. 1B). Three mutations (rmr6-1, rmr6-7, and rmr6-14) create premature nonsense codons, and one (rmr6-8) encodes a nonconservative amino acid substitution at a cysteine residue that is highly conserved among predicted RPD1 sequences (Fig. 1B and fig. S3). These molecular lesions strongly indicated that rmr6 encodes a Pol IV largest subunit.

Fig. 1.

(A) Diagram of the NRPD1a-like candidate gene model with lines connecting locations of transition-type lesions to resulting codon changes in the translated polypeptide represented in (B). (B) Peptide regions with >20% sequence similarity to conserved domains (A to H) of Pol II largest subunits (22) are shown in gray boxes, and the annotated domain G region (<20% sequence similarity) is identified with a dashed box. (C) Enriched sRNAs isolated from immature cobs of rmr6-1 homozygotes and heterozygous siblings. (D) Northern blot hybridization of enriched sRNAs from immature cobs with radiolabled doppia sequences. Stained transfer RNA bands (tRNA) below the blots are loading controls.

Additional molecular profiles also supported the assignment of RMR6 as a functional RPD1 ortholog. Arabidopsis showed a 14- to 15-fold reduction of 21- to 24-nt siRNAs in nrpd1a/nrpd1b double mutants relative to wild-type plants, and the accumulation of ∼94% of these RNAs was RPD1-dependent (19, 20). We fractionated sRNAs from immature cobs and noted by ethidium bromide staining intensity that the ∼24-nt RNA species was reduced an average of 82% (±5% SEM; n = 3 biological replicates) in rmr6-1 mutants, whereas the ∼21-nt RNAs were not affected (Fig. 1C). In Arabidopsis, the accumulation of 21-nt RNAs representing mature microRNAs (miRNAs) and trans-acting siRNAs (tasiRNAs) also appeared to be unaffected in PolIV and PolV mutants (7). Because Pol IV–dependent siRNAs are largely representative of transposon-like sequences in Arabidopsis (19, 20), and because of the effect rmr6 mutations have on cytosine methylation patterns 5′ of the Pl1-Rh coding region nearby and inside the proximal doppia-like transposon sequence (7), we specifically tested the effect of rmr6 mutations on the accumulation of doppia-like sRNAs. Northern blots showed that doppia-related sRNAs representing both strands were present in nonmutant plants but were undetectable in rmr6-1 mutants (Fig. 1D and fig. S5). Given the prevalence of doppia elements throughout the genome and the sequence arrangement within the terminal inverted repeats (12), it is unknown whether these sRNAs emanate from one or more elements or from any particular orientation. These molecular profiles, combined with the null-mutant phenotypes, indicate that rmr6 encodes a Zea mays Pol IV largest subunit, hereafter referred to as ZmRPD1.

All maize rmr loci were identified in mutant screens for plants unable to maintain paramutation-induced repression of alleles encoding color factors (for example, Pl1-Rh) (5, 6); however, rmr6 mutations also affect development (16). ZmRPD1 controls sex-determination fates by delimiting expression patterns of silkless1 (16), a gene responsible for protecting nascent ovules from a programmed-cell-death pathway (21). However, a reverse transcriptase polymerase chain reaction profile of rmr6 (fig. S6) showed that it is expressed in all rapidly dividing tissues throughout plant development, including the developing ear in which silkless1 action is required for ovule development (21). Lateral meristems are also often derepressed at nodal regions in rmr6 mutants, giving rise to ectopic organs (fig. S1) (16). Thus, Pol IV represents a molecular link between the regulation of developmentally important genes in maize and paramutation (6, 16).

Rice has undergone a duplication of the genes encoding RPD1 and RPE1 (14). However, we found only a single intact copy of an RPD1-encoding gene in all other available plant genomes, including Sorghum bicolor, a close relative of maize, the basal grass Brachypodium distachyon, and in the draft genome of maize, strongly indicating that the duplication of the RPD1-encoding gene is rice-specific. Furthermore, Southern blot analysis confirmed that rmr6 is singularly distinct in the maize genome (fig. S7), and in silico searches for rmr6-related DNA sequences in the regions syntenic with the rice RPD1-encoding duplicates were negative. The position of rmr6 within the maize B73 genome is also specific in relation to the S. bicolor genome.

A MAFFT ( based protein alignment was used to construct a parsimony tree (Fig. 2A) that agrees with the proposed evolutionary relationship of RPD1 to RPE1 and RPB1 (14) and confirms the conservation of catalytic center residues within RPD1 sequences (Fig. 2B) (13, 18). However, our alignment did not agree with previously published comparisons (18) over a ∼250–amino acid region of low sequence identity near the C terminus of the proteins, which includes domain G, one of the eight normally highly conserved RNA polymerase domains (22). We tested different alignment methods [MAFFT, T-COFFEE (, and MUSCLE (] which identified an RPD1 region that contains a deeply conserved backbone relative to all other polymerase largest subunits but which also contained different RPD1 sequences. We annotated this region as RPD1 domain G (Fig. 1B). In most DNA-dependent RNA polymerases, domain G contains a highly conserved motif known as the trigger loop (TL) that is both an interaction site of the potent inhibitor α-amanitin and a mediator of nucleotide selectivity during rapid transcript synthesis in yeast (23).

Fig. 2.

Phylogenomic analysis of RPD1. (A) Parsimony tree with bootstrap values generated from MAFFT alignment of full-length polymerase protein sequences from diverse plant species. (B) Excerpt from alignment used to generate the phylogenetic tree showing conserved DNA-dependent RNA polymerase domain D (33).

Arabidopsis Pol IV does not appear to have anyinvitro activity on isolated DNA substrates, and yet it is required for the in vivo accumulation of specific siRNA species (13). Cytological studies indicate that Pol IV acts upstream in the RdDM pathway in Arabidopsis (24), where it is hypothesized to generate single-stranded RNA transcripts of an aberrant nature from its target loci (13). If the maize Pol IV is a functional polymerase, Pol IV–dependent primary transcription should be lost in rmr6 mutants.

We performed nuclear run-on assays with nuclei isolated from husks of homozygous rmr6-1 mutants and heterozygous siblings in the presence or absence of α-amanitin. We detected no significant Pol IV–derived α-amanitin–insensitive transcription, as measured by comparing the ratio of α32P–uridine 5′-triphosphate (UTP) incorporated into run-on transcripts (–/+ α-amanitin) isolated from nuclei of nonmutant and homozygous rmr6-1 plants (Fig. 3A). Additionally, hybridization of labeled run-on RNAs from both α-amanitin–treated and –untreated nuclei from seedlings to slot blots with strand-specific RNA probes for a suite of maize sequence did not detect ZmRPD1-dependent transcripts (Fig. 3, B and C). The RNA probes used were homologous to potential ZmRPD1 templates represented by both the doppia and Mutator DNA transposons and the CRM and Cent-A retroelements (25). Although these repetitive sequences that are presumed targets of RdDM were proportionately less affected by α-amanitin treatment than the presumed Pol II–dependent transcript of the pigment biosynthesis factor A1 (Fig. 3C), they did not show the α-amanitin insensitivity of the Pol I–derived 45S precursor transcript (Fig. 3C) (26), nor did they show differences in abundance between rmr6 mutants and heterozygous wild-type siblings. Additionally, actinomycin-D treatment (Fig. 3B), which should abolish any DNA-dependent RNA polymerase activity (27), confirmed that all transcripts assayed were products of RNA synthesized from DNA templates.

Fig. 3.

(A) Average ratios (± SEM) of bulk 32P-UTP incorporation (–/+ α-amanitin) into RNA with isolated nuclei from husks of rmr6-1 homozygotes and +/rmr6-1 heterozygous siblings (n = 3 biological replicates). (B) Representative exposure of hybridized labeled run-on RNAs isolated from seedling nuclei of indicated genotype that were untreated (–) or treated with α-amanitin (α) or actinomycin-D (A-D). (C) Average ratios (± SEM) of quantified hybridization signal for run-on RNAs (–/+ α-amanitin) by densitometry for rmr6-1 homozygotes (gray bars) and heterozygous siblings (open bars) (n = 4 biological replicates for all probes except doppia and a1, for which n = 3 biological replicates). Hybridization signals were standardized to ubiquitin before calculation of –/+ α-amanitin ratio. (D) MAFFT alignment of ZmRPD1 domain G with region of distinct RPD1 sequence underlined.

Any Pol IV synthetic action should be α-amanitin–insensitive because the known α-amanitin interaction sites in RPB1 (bridge-helix and multiple TL residues) (23, 28) are divergent in RPD1 (31% identity/50% similarity across the bridge-helix and 11% identity/41% similarity across the TL) (Fig. 3D and fig. S3). Additionally, relative to comparisons between RPB1 and RPD1, the bridge helix and domain G of yeast Rpa1 and Rpc1 are more highly conserved to that of yeast Rpb1 (18), yet the cognate Pol I and Pol III complexes are both α-amanitin–insensitive (29). If maize Pol IV was enzymatically competent to synthesize RNA from either a DNA or nascent RNA template (13), we expected to detect α-amanitin–insensitive transcription, especially in light of the high-repeat content (∼70%) of the maize genome (30). Our results indicate that either Pol IV does not act as a RNA polymerase or that its relative amount of RNA synthesis is too low to detect by our comparisons. Extrapolations (17) of the data indicate that any potential ZmRPD1-dependent RNAs represent no more than ∼5% of the total labeled RNA in wild-type nuclei.

The reason for inefficient ZmRPD1 RNA polymerase activity may be tied to its altered domain G (Fig. 3D), which contains the TL in Pol II. Both α-amanitin treatment and a specific TL mutation decrease the synthetic rate (approximately 35-fold and 10-fold, respectively) of Saccharomyces cereviseae Pol II and increase the misincorporation incidence of both ribonucleotide and 2′-deoxyribonucleotides in in vitro assays (23). The altered domain G within the plant-specific RPD1 and RPE1 proteins may therefore be of considerable importance regarding any Pol IV enzymology.

We have shown that run-on transcription assays can be used as a powerful tool to study alternative polymerase function in plants. Although the biochemical function of RPD1 remains unclear, our results indicate that it acts as a component of a largely dysfunctional polymerase. Paradoxically, Pol IV is required for accumulation of the majority of ∼24-nt sRNAs, yet it provides no detectable RNA synthesis for genomic regions represented by those sRNAs in either maize (Fig. 3, B and C) or Arabidopsis (15). Pol II appears to be the primary, if not exclusive, polymerase for the repetitive features surveyed by our run-on transcription assays, including hypermethylated and repressed transposons such as doppia (7). These results, together with our previous analyses of ZmRPD1 function in facilitating paramutation (6), predict models of Pol IV action that can now be tested by combining biochemical and genetic approaches. Maize plants, which provide large amounts of specific tissues, represent an ideal system for further analysis of alternative polymerase action.

Although it remains unknown how ZmRPD1 is recruited to specific genomic features, its requirement for the accumulation of most 24-nt sRNAs indicates that it operates at repetitive features. Recruitment to transposon sequences proximal to Pol II templates could interfere with Pol II–dependent RNA synthesis, resulting in the production of abnormal Pol II transcripts. Local competition by RPD1 for Pol II holoenzyme subunits may interfere with Pol II function or cotranscriptional RNA processing; this idea is supported by the conservation of domain-specific RPD1 peptide sequences that are necessary for Rpb1 interaction with other Pol II subunits between yeast and maize (18). Alternatively, Pol IV may directly compete for Pol II templates and produce small amounts of RNA targeted for degradation pathways and sRNA amplification. Our analyses, combined with the recent characterization of TL function in yeast Pol II (23), indicate that such Pol IV RNAs might contain improper nucleotides that serve as a molecular tag for particular RNA degradation pathways. Such models involving abnormal RNA polymerase activities could apply to examples of paramutation outside the plant kingdom.

Our results point to a largely unknown facet of RNA metabolism by which homologous chromosomes can interact to affect gene regulation and heritable epigenetic change. Given that Rmr6 is required for both paramutation (6) and normalmaize development (16), its identification as ZmRPD1 illustrates that Pol IV plays a broader role in the biology of domesticated maize than in the eudicot Arabidopsis. This finding establishes a long-anticipated link between paramutation, developmental gene control, and heterochromatin function (31). The highly repetitive and heterochromatin-rich maize genome may influence the chromatin environment (and thus transcription) of a higher percentage of genes than in Arabidopsis, which has a smaller and less-repetitive genome. Identification of other genes influenced by rmr6 mutations thus promises new avenues for understanding these fundamental differences in plant development strategies and the role of paramutation in normal genome function.

Supporting Online Material

Materials and Methods

Figs. S1 to S7

Tables S1 to S4


References and Notes

View Abstract

Stay Connected to Science

Navigate This Article