Transcribed enhancer sequences are required for maize p1 paramutation

Abstract Paramutation is a transfer of heritable silencing states between interacting endogenous alleles or between endogenous alleles and homologous transgenes. Prior results demonstrated that paramutation occurs at the P1-rr (red pericarp and red cob) allele of the maize p1 (pericarp color 1) gene when exposed to a transgene containing a 1.2-kb enhancer fragment (P1.2) of P1-rr. The paramutable P1-rr allele undergoes transcriptional silencing resulting in a paramutant light-pigmented P1-rr′ state. To define more precisely the sequences required to elicit paramutation, the P1.2 fragment was further subdivided, and the fragments transformed into maize plants and crossed with P1-rr. Analysis of the progeny plants showed that the sequences required for paramutation are located within a ∼600-bp segment of P1.2 and that this segment overlaps with a previously identified enhancer that is present in 4 direct repeats in P1-rr. The paramutagenic segment is transcribed in both the expressed P1-rr and the silenced P1-rr′. Transcription is sensitive to α-amanitin, indicating that RNA polymerase II mediates most of the transcription of this sequence. Although transcription within the paramutagenic sequence was similar in all tested genotypes, small RNAs were more abundant in the silenced P1-rr′ epiallele relative to the expressed P1-rr allele. In agreement with prior results indicating the association of RNA-mediated DNA methylation in p1 paramutation, DNA blot analyses detected increased cytosine methylation of the paramutant P1-rr′ sequences homologous to the transgenic P1.2 subfragments. Together these results demonstrate that the P1-rr enhancer repeats mediate p1 paramutation.

The p1 gene encodes a myb-like transcription factor that regulates the biosynthesis of flavonoid-derived pigments (phlobaphenes) in mature tissues of maize (Lechelt et al. 1989;Grotewold et al. 1994).Accumulation of phlobaphene pigments in maize ears serves as a convenient indicator of p1 expression.Alleles of the p1 gene are designated by a 2-letter suffix with the first letter denoting pigmentation of the kernel pericarp (protective outer layer derived from the ovary wall) and the second letter indicating cob glume pigmentation.For example, alleles with red pericarp and red cob are denoted as P1-rr, alleles with red pericarp and white cob are referred to as P1-rw, alleles with white pericarp and red cob are named P1-wr, and alleles with white pericarp and white cob are indicated as p1-ww.
The P1-rr allele has a complex structure with the coding sequence flanked by compound direct repeats (Fig. 1a).Two long https://doi.org/10.1093/genetics/iyad178Investigation direct repeats of 5.2 kb (black rectangles) are located immediately upstream and downstream of the P1-rr coding sequence and are juxtaposed with 1.2-kb direct repeats (hatched rectangles) (Fig. 1a).The upstream-most 1.2-kb repeat is truncated and interrupted by a 1.6-kbhAT-family transposon (Goettel and Messing 2010), while the downstream 1.2-kb repeats overlap with the exons 3 and 4 of P1-rr (Grotewold et al. 1991;Goettel and Messing 2010).A fractured Mu-like transposable element (fMULE; Fig. 1b) is present within 1.2-kb repeats, 1 copy within the upstream 1.2-kb repeats and 2 copies in the downstream 1.2-kb (a) (b) Fig. 1.Structure of the P1-rr locus and transgenic P1.2 subfragments.a) The P1-rr gene map.Starting at the top, the location of Ac transposon insertion that disrupts P1.2 distal enhancer and results in reduced pericarp pigmentation (Sidorenko et al. 2000) is shown by a triangle.Coordinates of the P1.2 and Pb fragments (double-headed arrows) are shown relatively P1-rr transcription start site.The major P1-rr transcript containing exons 1, 2, and 3 (E1, E2, and E3) and an alternative P1-rr transcript containing exons 1, 2, and 4 (E1, E2, and E4) are indicated by black boxes with introns shown as thin lines.A 1.6-kb hAT-like transposable element that disrupts the upstream-most 1.2-kb repeat is indicated by a medium gray box.Copies of the p1 cob glume enhancer (Zhang and Peterson 2005) are shown as black boxes.Along the P1-rr map, the 5.2-kb-long direct repeats (black rectangles) overlapping with smaller 1.2-kb direct repeats (hatched boxes), and SalI restriction sites are indicated.Bent arrow shows P1-rr transcription start site at position +1; ATG indicates the P1-rr translation start site.Below the map: double-headed arrows indicate the location of the SalI-delimited upstream and downstream copies of the P1.2 fragment, which are 99.84%identical (Su et al. 2020).b) Schematic drawings of the transgenic P1.2 subfragments.Locations of the major P1.2 subfragments and SalI restriction sites are shown on the top of b (Lechelt et al. 1989).The truncated copy of 13 (Δ13) lacks the downstream SalI (Supplementary Fig. 1b).Locations of p1 cob glume enhancer (Zhang and Peterson 2005) are shown as black rectangles.A fractured MULE ( fMULE) is shown as a light gray rectangle (Goettel and Messing 2010;Goettel and Messing 2013a).The fragment of P1-rr′ with increased CHH methylation reported by Sekhon et al. 2012 is shown as checkered box.The SalI, SacI, and PstI restriction sites are indicated along the thick black line.The P1.2 subfragments are diagrammed as rectangles below the restriction map.Within each construct diagram, black rectangles indicate sequence of the p1 cob glume enhancer, light gray rectangles diagram fMULE, and open rectangles show sequence unique to p1 locus.For simplicity, the basal P1-rr promoter and 5′UTR that were included in all transgene constructs (Materials and methods) are not shown.Construct names are indicated to the right of the construct diagrams.Length of the P1.2 subfragments are indicated in base pairs (bp) before each construct name.Detailed sequence information for the P1.2 subfragments is shown in Supplementary Fig. 1a.
repeats (Goettel and Messing 2013a).The 1.2-kb direct repeats contain SalI restriction sites (Lechelt et al. 1989) that delimit the upstream P1.2 (located ∼5-kb upstream of P1-rr transcription start site, at −6,110 to −4,842 in AF209212.1)and the downstream P1.2 (overlapping with the alternatively spliced p1 transcript; Fig. 1a).Insertion of Ac transposon into the upstream copy of the P1.2 sequence (Fig. 1a) of P1-rr results in reduced ear pigmentation, indicating the importance of this upstream element for p1 expression (Athma et al. 1992;Moreno et al. 1992;Sidorenko et al. 2000).Subsequent transient and stable transgenic plant assays confirmed enhancer function of the P1.2 fragment (Sidorenko et al. 1999(Sidorenko et al. , 2000)).Comparing phenotypes, RNA expression patterns, and the upstream regulatory sequences of P1-rr and P1-rw alleles identified a 386-bp sequence within the upstream-most 1.2-kb repeat of P1-rr as a cob glume enhancer (Fig. 1a and b) (Zhang andPeterson 2005, 2006).Together, these studies of structural variation of endogenous alleles and functional tests in transgenic experiments demonstrated the importance of the P1.2 enhancer fragment in P1-rr regulation.
Paramutation at the p1 gene was first observed in experiments that studied epigenetic interactions between the endogenous P1-rr allele and transgenes carrying fragments of P1-rr regulatory region (Sidorenko and Peterson 2001).When introduced as a transgene, the P1.2 enhancer, fused to the basal P1-rr promoter (Pb) and GUS gene (P1.2b::GUS),caused silencing of the endogenous P1-rr allele, resulting in reduced pigmentation and variable patterning of pericarp and cob glumes.Additional tests demonstrated that Pb was not required and that P1.2 was sufficient for paramutation (Sidorenko and Chandler 2008).By convention with paramutation nomenclature, silenced P1-rr epialleles are termed P1-rr′, with the prime mark signifying the epigenetically silenced state resulting from paramutation.The silenced P1-rr′ state was heritable and caused secondary silencing (secondary paramutation or paramutagenicity) of a naïve (no previous exposure to a transgene) P1-rr allele in the absence of the inducing P1.2 transgene (Sidorenko and Peterson 2001;Sidorenko and Chandler 2008).Molecular analyses revealed that the transgene-induced silenced state was associated with increased DNA methylation of the endogenous 1.2-kb repeats and reduced P1-rr transcription (Sidorenko and Peterson 2001;Sidorenko and Chandler 2008).The frequency of silencing induced by the P1.2-containing transgenes varied between independent transgenic events, as did the heritability and secondary paramutagenicity of the newly induced P1-rr′ states (Sidorenko and Peterson 2001;Sidorenko and Chandler 2008).Generally, progeny derived from strongly silenced P1-rr′ epialleles were more likely to remain heritably silenced and to exhibit secondary paramutation than progeny from weakly silenced P1-rr′ alleles, which frequently reverted back to the fully pigmented red P1-rr state in the absence of the inducing transgene (Sidorenko and Peterson 2001).These observations indicated that newly established P1-rr′ silencing was metastable, although stable lineages could be selected.This is similar to the properties of alleles participating in paramutation at pl1 and r1 loci where metastable states have been reported (Styles and Brink 1966;Brink et al. 1968;Styles et al. 1973;Hollick et al. 1995) but distinct from b1 paramutation in which silencing and secondary paramutation are extremely stable (Coe 1959(Coe , 1966)).
In addition to transgene-induced paramutation of P1-rr (Sidorenko and Peterson 2001;Sidorenko and Chandler 2008), spontaneous epialleles of P1-rr with patterned pericarp and red cob designated as P1-pr have been described.The 2 independently isolated P1-pr epialleles (Das and Messing 1994;Sekhon et al. 2012) are phenotypically indistinguishable from each other and from P1-rr′ induced by the P1.2 transgene, but they differ in the stability of the silenced states and their paramutation characteristics.Similar to P1-rr′ (Sidorenko and Peterson 2001), the P1-pr isolate characterized by Das and Messing (1994) is mostly stable, although low-frequency reversions were reported (Das and Messing 1994;Goettel and Messing 2013a).In contrast, the P1-pr TP isolate is very stable with no revertants observed in ∼1,000 plants examined (Sekhon et al. 2012).Interestingly, the P1-pr isolated by Das and Messing is highly paramutagenic (Goettel and Messing 2013b), while P1-pr TP is not paramutagenic and behaves as a recessive allele in crosses with P1-rr (Sekhon et al. 2012).Cytosine methylation assays showed that P1-rr′ and both P1-pr isolates have similar high densities of symmetric CG and CHG (H is any base but G) methylation in the region of the 1.2-kb fragment.However, there is a distinct difference in nonsymmetric CHH methylation; the paramutagenic P1-rr′ and P1-pr had ∼10% and ∼5% CHH methylation, respectively (Sekhon et al. 2012;Goettel and Messing 2013a), while no detectable CHH methylation was observed for nonparamutagenic P1-pr TP (Sekhon et al. 2012).Because CHH methylation depends on presence of small RNAs (sRNAs) involved in RNA-dependent DNA methylation (RdDM) (Matzke and Mosher 2014), the difference in nonsymmetric CHH methylation was proposed to be a distinguishing molecular mark between the paramutagenic and nonparamutagenic P1-rr epialleles (Sekhon et al. 2012).
Further insights into the molecular mechanisms involved in p1 paramutation were obtained in genetic experiments with RdDM mutants that affect paramutation at other maize loci.The first gene shown to be required for p1 paramutation was mop1 (mediator of paramutation 1) (Dorweiler et al. 2000;Alleman et al. 2006).The Mop1 gene is allelic to Rmr6 (required to maintain repression 6) (Erhard et al. 2009) and encodes an RNA-dependent RNA (RDR) polymerase orthologous to Arabidopsis RDR2 (Alleman et al. 2006).The mop1-1 mutation disrupted the establishment of p1 paramutation, but in contrast to b1 paramutation in which silencing was reversed in the first generation (Dorweiler et al. 2000), effects on the maintenance of P1-rr′ silencing were delayed; several generations of exposure to the homozygous mop1-1 mutation were required to restore pigmentation of P1-rr′ (Sidorenko and Chandler 2008).Slow reactivation of P1-rr′ in mop1 mutant background correlated with modest increase in P1-rr transcript levels (∼40%), slight decrease of repressive H3K9me 2 histone mark (∼30%), and reduction of CHH DNA methylation from 7.1 to 3.8%, while CG and GHG methylation was not significantly different (Wang et al. 2017).The second gene shown to be required for p1 paramutation is mop2 (Sidorenko et al. 2009).This gene is allelic to Rmr7 (Stonaker et al. 2009) and encodes one of 3 paralogs of the second largest subunit of RNA polymerases IV and V [Pol-IV/V; also referred to as NRP(D/E)2a (Haag et al. 2014)].In the heterozygous state, the Mop2-1 mutation disrupted the establishment of pl paramutation, and this dominant mutant behavior was similar to the results with b1 paramutation (Sidorenko et al. 2009).However, in contrast to b1 paramutation, the Mop2-1 allele did not affect maintenance of p1 paramutation (Sidorenko et al. 2009).The effects of mutations in mop1 and mop2 genes that are orthologous to genes in the Arabidopsis RdDM pathway demonstrates the involvement of RNA-mediated regulation in p1 paramutation (Sidorenko and Chandler 2008;Sidorenko et al. 2009).In contrast to the mop1-1 and Mop2-1 mutations, the dominant Ufo1-1 (Unstable factor for orange1) mutation (Wittmeyer et al. 2018) disrupts previously established P1-rr′ silencing in the first generation of exposure to the mutation (Sekhon et al. 2012) suggesting a critical requirement for UFO1 in maintaining P1-rr′ silencing.At the molecular level, Ufo1-1 mutation arose via a CACTA transposon insertion and increased expression of a Poaceae-specific protein with no annotated function (Wittmeyer et al. 2018).While the mechanism by which Ufo1-1 derepresses silenced P1-rr′ is unknown, the presence of Ufo1-1 resulted in decreased cytosine methylation within a 443-bp subfragment of the P1.2 enhancer in all sequence contexts (CG, CHG, and CHH) and reduced accumulation of the repressive H3 lysine 9 di-methylated histone modification (H3K9me2) (Sekhon et al. 2012).Together, these genetic studies show that, similar to other cases of maize paramutation (Dorweiler et al. 2000;Alleman et al. 2006;Erhard et al. 2009;Stonaker et al. 2009), establishment of p1 paramutation depends on RNA-mediated silencing mechanisms, while maintenance of P1-rr′ silencing has distinct genetic requirements that would benefit from additional studies.These additional studies may include factors that have been shown to regulate pl1 and b1 paramutations, including but not limited to PICKLE-like chromodomain DNA-binding 3 (CHD3) protein that is required to maintain pl1 paramutation but is dispensable for maintenance of b1 paramutation (Deans et al. 2020) and AGO104 that disrupts b1 paramutation (Aubert et al. 2022).
The aim of the present study is to further define the minimal sequences within the P1.2 fragment required for paramutation.We designed a series of P::GUS constructs containing subfragments of P1.2 (Fig. 1b) and stably transformed these constructs into maize.The resulting transgenes were tested for their ability to induce silencing of P1-rr, and the heritability and secondary paramutagenicity of the resulting P1-rr′ states were characterized.These experiments identified a minimal paramutagenic sequence of ∼600 bp within P1.2 and further documented the involvement of cytosine methylation, transcription, and sRNA accumulation in paramutagenic silencing.

Plasmid construction
Constructs used in the study were assembled using standard molecular cloning techniques in a vector described in Sidorenko et al. (1999).All constructs contained the Pb fragment (−235 to +326, AF209212.1)including the basal P1-rr promoter and 5′UTR of the P1-rr gene, which were fused to the maize AdhI gene intron I, the β-glucuronidase (GUS) reporter gene, and the potato Proteinase Inhibitor II (PinII) terminator.A diagram of the P1.2 subfragments used in this study is shown in Fig. 1a and b with detailed information in Supplementary Fig. 1.The P1.2 subfragments 13, 14, and 15 are the same as P1-rr locus fragments 13, 14, and 15 described in Lechelt et al (1989).The P1.2b::GUS and Pb::GUS constructs were previously described (Sidorenko and Chandler 2008).For simplicity, in this study, P1.2b::GUS is referred to as P1.2, and Pb::GUS is referred to as Pb.

Plant transformation
Transgenic plants production was essentially as in Sidorenko and Chandler (2008).Briefly, transgenic maize plants were produced at the Iowa State University Transformation Facility (ISU PTF) using a biolistic bombardment protocol.Plasmids of interest and the pBAR184(−) selectable construct were cobombarded into Hi-II immature embryos (Armstrong 1994;Frame et al. 2000).Hi-II carries the p1-ww allele that does not participate in paramutation.The T 0 transgenic plants were crossed with a P1-rr tester (see description of genetic stocks) to produce seeds for further testing (Supplementary Fig. 2).

Molecular analyses of transgenic callus and maize transgenic plants
Transgenic callus clones were screened for the presence of transgenic constructs of interest using DNA blot analyses as described by Sidorenko and Chandler (2008).Briefly, genomic DNA was digested with restriction enzymes, electrophoresed, blotted, and hybridized with a 32 P-labeled 788-bp NcoI-AlwNI fragment from the GUS gene coding region.Only GUS-positive callus clones were regenerated.Transgenic lines were evaluated in T 0 and/or T 1 generations as follows: leaf genomic DNA was digested with restriction enzymes that released the entire P::GUS cassette; digested DNA was electrophoresed, blotted, and hybridized with 32 P-labeled GUS and/or 15 probes; and the number of discrete bands were counted and reported in Supplementary Fig. 3. Additionally, transgene copy number was estimated by comparison of DNA gel blot hybridization signals of transgene and endogenous P1-rr bands with probe 15 using Phoretix 1D software package (Sigma-Aldrich).For this, bands were detected as recommended by software developer and background subtracted.Lanes containing samples from nontransgenic plants were used as controls to estimate probe 15 copy number in lanes containing transgenic samples.Endogenous probe 15 copies were subtracted from the total estimated probe 15 copy number, and the resulting estimated probe 15 copy numbers are reported in Supplementary Fig. 3.

Genetic stocks
The P1-rr stock used in this study was the standard P1-rr4B2 allele (Grotewold et al. 1991).Except for initial crosses with T 0 transgenic plants that were conducted at the ISU PTF, large-scale genetic experiments with transgenic plants employed isolation fields where transgenic families were detasseled and their ears were allowed to open pollinate with nontransgenic testers planted in alternating rows.Because isolation fields were in Arizona, we developed hybrid P1-rr and p1-ww tester stocks that performed well in the dry and hot weather.For the hybrid P1-rr stock, we crossed the near isogenic inbred lines of P1-rr4B2[4Co63] and P1-rr4B2 [W23].For the hybrid p1-ww stock, we crossed p1-ww[4Co63] and p1-ww[A619] inbred lines.The p1-ww[4Co63] allele is neutral to paramutation and carries a single copy of fragment 15 (Sidorenko and Peterson 2001;Sidorenko and Chandler 2008;Goettel and Messing 2013b).Similar to p1-ww[4Co63], the null p1-ww[A619] allele carries a single copy of fragment 15 (Szalma et al. 2005;Morohashi et al. 2012) and was not observed to participate in paramutation (Sidorenko, unpublished).

Genetic tests with transgenic plants
Diagrams of crosses used in this study are shown in Supplementary Figs. 2, 4, and 6.In these diagrams, presence of a transgene is indicated as TR, and asterisk is used to indicate P1-rr that was exposed to the transgene.To identify transgenic plants in segregating families, all plants were screened for tolerance to Ignite herbicide (Bayer) using the leaf painting assay.For this, a permanent black marker was used to mark the location of herbicide application on a fully expanded maize leaf near the leaf tip.Herbicide solution (1% v/v) was applied with a sponge next to the marker line.Three days after application, herbicide tolerance was visually scored; plants with leaf damage were scored as sensitive, while plants with no damage as tolerant.To confirm the absence of transgenes in the herbicide-sensitive plants, DNA from these plants was subjected to PCR amplification with GUS primers (forward 5′-GCGTCGGCATCCGGTCAGTGGC-3′ and reverse 5′-TCTGCCGTTTCCAAATCGCCGC-3′) and/or DNA slot blot analyses with a 32 P-labeled NcoI-AlwNI GUS fragment probe.To distinguish P1-rr/P1-rr from P1-rr/p1-ww genotypes in the secondary paramutagenicity tests, DNA of all nontransgenic plants were subjected to PCR amplification and genotyping by sequencing.The sequences of the primers used for genotyping were obtained from R. Sekhon (forward 5′-GTCGCGTGGGTCTTCG TTCAG-3′ and reverse 3′-ATGCAAGTGACAAATTATCATGATG GCA-5′).Resulting 452-bp PCR product spanned a simple sequence repeat polymorphism (Supplementary Fig. 1a) that was used for genotyping.In this assay, the P1-rr/P1-rr plants produced uninterrupted sequence traces, while P1-rr/p1-ww plants displayed easily identifiable shift in sequencing traces after the polymorphism (GTGTGT in P1-rr and GTGTGTGT in p1-ww; Supplementary Fig. 1a).PCR products were sequenced at the University of Arizona Genetics Core Facility using the reverse primer described above.

Statistical analysis
To compare frequency of silenced ears of each construct with that of the control P1.2 construct, we carried out 2 sample tests for equality of proportions with Yates' correction for continuity.The test hypotheses to compare frequency of silenced ears are as follows: H 0 , The frequency of silenced ears of a construct with a fragment of P1.2 is equal to frequency of silenced ears of the construct with complete P1.2; H A , The frequency of silenced ears of a construct with a P1.2 fragment is not equal to frequency of silenced ears of the construct with complete P1.2.The P-values obtained from the hypothesis testing were adjusted using Bonferroni-Holm method for multiple comparison correction.

DNA blot analyses of cytosine methylation
DNA blot analyses were conducted using leaf genomic DNA extracted by a CTAB protocol (Saghai-Maroof et al. 1984;Wise and Schnable 1994).Restriction enzymes not sensitive to cytosine methylation (EcoRI and KpnI) were used to assess the transgene locus structure, while enzymes sensitive to CpG (SalI) and CpNpG (PstI) cytosine methylation were used to assess methylation of the endogenous P1-rr allele.Standard DNA blot protocol (Sidorenko et al. 2000;Sidorenko and Peterson 2001) involving DNA digestion, electrophoresis, transfer to Nylon membrane, hybridization with 32 P random labeled probe, washing, and image acquisition was used in this study.

Nuclear run-on analysis
Nuclear run-on analysis was conducted using fresh tissues harvested from field grown plants.Young inner leaf sheaths were collected from plants at the V7 growth stage.Young inner husks were collected from developing ears before silk extrusion.Immature pericarps were collected from kernels 20 days after pollination.To extract nuclei, 5 g of a respective tissue was ground in liquid nitrogen and resuspended in 30 ml of ice-cold extraction buffer (Sidorenko and Chandler 2008).The slurry was filtered through Miracloth (Calbiochem) and 53 µm nylon mesh.The filtrate was transferred to centrifuge tubes and centrifuged for 15 min at −10° C and 6,000 g.The supernatant was removed, and nuclei were resuspended using a soft paint brush, washed with 20 ml of the wash buffer, pelleted again, and resuspended in 2 ml of the wash buffer (Lisch et al. 2002;Carey et al. 2004).The remaining run-on reaction was as described by Dorweiler et al. (2000).Total RNA from nuclei was extracted using TRIzol reagent (Life Technologies) according to the manufacturer recommendations.Hybridization was as described previously (Dorweiler et al. 2000).To prepare RNA probes for the 13, 15α, and 15β subfragments and the 18S control, PCR-amplified fragments containing the T3 promoter were used as templates.The in vitro transcription reactions were performed using T3 RNA polymerase as recommended by the manufacturer (New England BioLabs).The maize Ubiquitin2 (Ubi2) RNA probe was prepared using SP6 RNA Polymerase and the linearized pCA210 plasmid (Christensen et al. 1992).A DNAseI digestion was used to remove DNA template from the in vitro synthesized RNA probe preparations.To prepare slot blots, 200 ng of denatured RNA was immobilized on GeneScreenPlus Nylon membrane (PerkinElmer).After hybridization and washing, blots were scanned using a BioRad Pharos scanner, and pixel volumes were determined using the accompanying software package.After subtracting the background signal, pixel volumes per cm 2 were normalized to the pixel volumes obtained for Ubi2.In experiments with α-amanitin, pixel volumes were not normalized as transcription of all genes assayed in this experiment was expected to be affected.

Transgenes carrying P1.2 subfragments had complex multicopy transgene insertions
Transgenic plants were produced using a biolistic bombardment protocol in which P::GUS constructs were cobombarded with the pBAR184(−) selectable construct (Armstrong 1994;Frame et al. 2000).As expected for biolistic cotransformation method, transgenic events exhibited multicopy P::GUS transgene insertions with a wide variation of copy numbers (2 to 163 copies; Supplementary Fig. 3).A combination of herbicide leaf paint test and molecular analyses were used to evaluate transgene segregation in T 1 plants (Materials and methods).Results revealed that roughly 1/3 of the transgenic events (24 out of 77) exhibited deviations from the expected 1 herbicide-tolerant:1 herbicide-sensitive segregation (Supplementary Fig. 3).Among events with skewed segregation, 13 events had higher than expected and 11 events lower than expected frequencies of herbicide-tolerant plants.
Enhancer sequences mediate p1 paramutation | 5 Higher than expected frequency of herbicide-tolerant plant could be a result of multiple unlinked transgene insertions (Spencer et al. 1990;Register et al. 1994), while lower than expected frequency of herbicide-tolerant plants could occur when the selectable marker becomes silenced (Kumpatla et al. 1997;Rajeevkumar et al. 2015).While we were not able to follow up on every case of deviation from the expected 1:1 segregation, we identified 7 cases of unlinked transgene insertions that increased number of HT plants (indicated by red stars in Supplementary Fig. 3).We also identified 93 plants sensitive to herbicide and carrying P::GUS transgenes distributed among 24 transgenic events (Supplementary Fig. 3), indicating that silencing of selectable marker occurred sporadically in many transgenic events.
Because the relationship between transgenic event structure and its ability to induce P1-rr silencing is of interest, we explored correlation analyses using data in Supplementary Fig. 3.We detected positive correlation between number of transgenic bands and P1-rr silencing for 14 + 15a and Pb constructs (R 2 0.57, P = 0.004 and R 2 0.65, P = 0.05, respectively), while the other constructs showed no significant correlation (not shown).Using estimated copy number for correlation analyses did not reveal significant correlation for any of the constructs (not shown).These results indicate that the highly variable and rearranged transgene structures generated by biolistic bombardment are not suitable for evaluation of copy number impact on P1-rr silencing.
In conclusion, our results of transgenic events characterization are similar to the previously published results reporting biolistic bombardment to produce complex, multicopy transgene insertions that can exhibit sporadic selectable marker silencing.Despite these limitations, transgenic plants served as the essential enabling tool in functional dissection of sequences within the required for p1 paramutation P1.2 fragment.

Sequences within subfragment 15 of the P1.2 enhancer are required for transgene-induced P1-rr silencing
To assay for silencing of P1-rr, primary T 0 transgenic plants homozygous for p1-ww and hemizygous for a transgene (TR/-) were crossed with the nontransgenic homozygous P1-rr tester and progeny plants evaluated for silencing (Materials and methods).In these crosses, exposure of naive P1-rr to a transgene is signified by asterisk (Supplementary Fig. 2).As expected, all nontransgenic T 1 P1-rr ears (not shown) had dark red pericarp and cob pigmentation (Fig. 2a).This result is in agreement with prior observations that P1-rr is a stable allele and does not undergo appreciable spontaneous silencing (Das and Messing 1994).In contrast, many transgenic plants produced ears with reduced P1-rr pigmentation (Fig. 2b-j).The P1.2 transgene construct, containing the complete 1,269-bp enhancer fragment and used as a positive control for silencing (Sidorenko and Chandler 2008), induced P1-rr silencing in all 9 tested events (Fig. 3; Supplementary Fig. 3).The average silencing frequency, calculated as average of silencing frequencies of individual events, was 72%, and the majority of affected plants exhibited phenotypically strong silencing (Supplementary Fig. 3) similar to that shown in Fig. 2d, e, and j.
To further localize sequences required for silencing, 9 constructs containing subfragments of P1.2 were tested.Subfragments of P1.2 (Fig. 1b) were developed (Materials and methods) based on previously characterized P1-rr restriction fragments 13, 14, and 15 (Lechelt et al. 1989).One of the constructs (13 + 14) contained 13 and 14 subfragments and lacked 15.In contrast to strong and frequent silencing induced by P1.2, this construct exhibited only phenotypically weak P1-rr silencing (Fig. 2b and c) only in 1 out of 5 tested events, resulting in average frequency of 11% (Fig. 3; Supplementary Fig. 3).Performance of this construct was similar to that of the Pb control construct (Sidorenko and Chandler 2008) that contained only the P1-rr basal promoter and 5′UTR and exhibited phenotypically weak silencing in 3 out of 6 tested events with average frequency of 11% (Fig. 3; Supplementary Fig. 3).This result showed that sequences within 13 and 14 were not sufficient to induce frequent and phenotypically strong silencing and that subfragment 15, absent from this construct, is essential for transgene-induced silencing of P1-rr.Indeed, the critical role of 15 was confirmed by constructs containing partial or complete fragment 15 sequences (Fig. 3; Supplementary Fig. 3).Three of these constructs (ΔΔ14 + 15, 13 + 15, and 15) contained the entire fragment 15; all tested events induced phenotypically strong P1-rr silencing (Supplementary Fig. 3), and average frequencies of silencing (Supplementary Fig. 3;77,84,and 69%,respectively) were not statistically different from the full-length P1.2 (72%).Five additional constructs contained one of 2 subfragments of 15 (15α or 15β).For these constructs, not all events induced P1-rr silencing, and average silencing frequencies were overall lower (18-52%) than those for the complete P1.2 control construct (Fig. 3; Supplementary Fig. 3).Together, these results show that the 864 bp of fragments 13 and 14 were not sufficient,  2).The average frequency of silencing for a construct was calculated as sum of frequencies of silencing for each event divided by the number of tested events (Materials and methods).Statistically significant differences relative the P1.2 control are indicated by 2 stars for P < 0.001 and 3 stars for P≤0.0001.P-values were calculated by 2-sample test for equality of proportions with Yates' correction for continuity, followed by Bonferroni-Holm adjustment for multiple comparisons (Materials and methods).The box plot on the right shows silencing frequencies for individual events with each of the dots sized proportionally to the number of transgenic plants assayed for that event.Light-and dark-shaded areas of the box plot indicate 2 middle quartiles with the dividing line indicating median frequency of silencing.The whiskers indicate 1.5 times the minimum and maximum of the respective interquartile range.Additional details regarding transgene segregation, results of molecular analyses, and frequencies of strong and weak P1-rr silencing are in Supplementary Fig. 3. while the 409-bp fragment 15 was necessary and sufficient for phenotypically strong and frequent silencing of P1-rr.

Sequences required for high heritability of transgene-induced P1-rr silencing are located within the 13 and 15 subfragments
Previous studies have shown that silencing of P1-rr induced by a P1.2 transgene remained heritable even after the transgene was removed by genetic segregation (Sidorenko and Peterson 2001;Sidorenko and Chandler 2008).To assay the heritability of the P1-rr*-silenced states induced by subfragments of P1.2, transgenic plants were crossed with the nontransgenic p1-ww tester (Supplementary Fig. 4).Progeny plants containing P1-rr* and lacking a transgene were identified (Materials and methods).Because previous results with the P1.2 transgenes demonstrated that the heritability of P1-rr silencing was low in progeny of weakly silenced P1-rr ears (Sidorenko and Peterson 2001), heritability of P1-rr* silencing was tested only in progeny of strongly silenced ears (similar to those shown in Fig. 2d, e, and h-j).In this way, we tested heritability of transgene-induced P1-rr silencing for 7 constructs; the results are shown in Fig. 4 and Supplementary Fig. 5.For the control P1.2 construct, silencing was heritable in all 7 tested transgenic events; on average 83% of the nontransgenic ears per event were silenced.Similarly, P1-rr silencing induced by the 13-containing constructs, 13 + 15 and 15β + Δ13, was highly heritable for all tested events; average frequency of silencing per event was 80 and 74%, respectively.In contrast, not all events of constructs lacking 13 (ΔΔ14 + 15, 15, Δ14 + 15α, and 15β) exhibited heritable P1-rr silencing.For these constructs, the average frequency of heritable silencing was lower than that for the P1.2 control (26%, 31%, 7%, and 2%, respectively).Furthermore, the majority of silenced ears (Supplementary Fig. 5) exhibited significant amounts of residual ear pigmentation similar to that in Fig. 2b, c, f, and g.Thus, these results showed that, while not required for establishing silencing of P1-rr, the 13 subfragment was required in addition to subfragment 15 to confer high heritability of P1-rr silencing in the absence of the inducing transgene.

Subfragments 13 and 15 contain sequences required for strong secondary paramutation
Secondary paramutation, i.e. the ability of a silenced P1-rr* to paramutate a naïve P1-rr allele in the absence of the original silencing transgene, is a defining characteristic of paramutation.Secondary paramutation tests were conducted for 8 constructs using families derived from the strongly silenced P1-rr* ears (similar to those in Fig. 2d, e, and h-j).To assay secondary paramutation, transgenic plants were crossed with the nontransgenic P1-rr tester (Supplementary Fig. 6), and progeny plants homozygous for P1-rr and lacking a transgene were identified and scored (Materials and methods).Results are summarized on a construct level in Fig. 5 with data for individual events shown in Supplementary Fig. 7.For the control P1.2 construct, secondary paramutation was assayed for 4 transgenic events: the average frequency of silencing was 27%.Among constructs carrying subfragments of P1.2, multiple independent transgenic events were tested for 13 + 15 (2), 15 (3), and 15β + Δ13 (3).One construct, 13 + 15, gave an average frequency of silencing greater than that of P1.2 control (60%), while 2 constructs, 15 and 15β + Δ13, exhibited frequencies of silencing not significantly different from P1.2 control (19 and 29%, respectively).For the remaining 4 constructs (ΔΔ14 + 15, 13 + 14 + 15α, Δ14 + 15α, and 15β), we were able to test only 1 event per construct.From these constructs, only the 15β construct showed result significantly different from P1.2 control, with 0% silencing.The Δ14 + 15α construct also showed 0% silencing, but in this case, only 13 plants were tested, rendering this result not statistically significant.Finally, the ΔΔ14 + 15 and the 13 + 14 + 15α gave moderate frequencies of silencing (18 and 43%, respectively) that were not significantly different from the P1.2 control.Based on these results, we conclude that sequences required and sufficient for paramutation are located within 13 and 15 subfragments.In accordance with established paramutation terminology, the P1-rr* epialleles silenced by the constructs that induced heritable and secondary paramutagenic silencing will be further designated as P1-rr′.

Distinct "blushed" ear phenotype preferentially induced by constructs lacking subfragment 13
Initial experiments showed that the P1.2 transgenes elicited a variety of P1-rr′ ear phenotypes, ranging from light red to very pale kernel pericarp and cob glumes (Fig. 2b-e).Kernels on silenced ears often exhibited prominent red stripes emanating from the silk scar region (Fig. 2d and e), a phenotype termed "dark crown" by Emerson (Emerson 1917).Analyses of additional transgenic events containing P1.2 (Sidorenko and Chandler 2008) also identified fully silenced P1-rr′ states with colorless pericarp and cob glumes (Fig. 2j).Interestingly, in this study using transgenes carrying subfragments of P1.2, we observed a distinct pigmentation pattern of "blushed" pericarp, weak cob glume color, and absence of dark crown kernels (Fig. 2f-i).Closer examination of hand sections from representative kernels of patterned and blushed phenotypes revealed differences in pigment deposition within the pericarp.In dark crown patterned kernels, sections through the red stripe revealed pigment accumulation in the crown region and an outer cell layer of pericarp from the kernel gown (side) (Fig. 2k).In contrast, sections of blushed kernels exhibited colorless crown and pigmented inner pericarp cell layers of the kernel gown (Fig. 2l).These distinct pericarp pigmentation patterns coincide with spatial location of epidermal LI and subepidermal LII pericarp cell lineages (Dellaporta et al. 1991) and suggests a possibility that the dark crown pattern is associated with a greater silencing of the inner LII pericarp cell layers, while the blushed phenotype is associated with a greater silencing of the outer LI cell layers.
In summary, the distinct blushed ear phenotypes were preferentially induced by constructs lacking subfragment 13.Moreover, the blushed ear phenotypes were unstable and frequently reverted to red pericarp and cob pigmentation.These observations are consistent with the idea that the blushed ear phenotype results from silencing being predominantly localized to the pericarp LI cell layers.Because the female megasporocyte develops from the LII cells (Dellaporta et al. 1991), a lack of silencing in the LII may account for the low heritability of silencing and frequent reversion of the blushed ear phenotype.
Consistent with prior results, active P1-rr and heritably silenced P1-rr′ epialleles exhibit distinctly different DNA methylation patterns as assessed by SalI and PstI restriction enzyme digestion and hybridization with probe 15 (Fig. 6a; Supplementary Fig. 9b  and c).For the active P1-rr allele (first lanes), SalI produces bands of 3.3, 2.7, and 1.2 kb (doublet), while PstI produces bands of 10.3, ∼5.0, 2.9, 1.2-kb doublet, and 0.7 kb.The heritably silenced  9c, 10c, and 11c).a) Map of the endogenous P1-rr gene locus.P1-rr exons (black boxes) and introns (thin lines) (Grotewold et al. 1991) and location of probe 15 (gray boxes) (Lechelt et al. 1989)  To summarize, these results revealed patterns of P1-rr′ methylation associated with both strength of silencing and degree of heritability.Strong and highly heritable P1-rr silencing induced by the P1.2 and 13 + 15 transgenes correlated with complete methylation of SalI and PstI sites located within the transgenehomologous sequences.However, strongly silenced but weakly heritable P1-rr silencing induced by ΔΔ14 + 15, Δ14 + 15α, and 15β transgenes correlated with partial methylation of SalI and PstI sites.In these cases, partially methylated or not methylated restriction sites were located at the ends of the homology to transgene fragments.These results are reminiscent of previous studies showing that methylation of integrated transgenes is restricted to the transgene and does not spread to adjacent sequences (Day et al. 2000).

Sequences mediating p1 paramutation are transcribed in both active P1-rr and silenced P1-rr′
Transcription by DNA-dependent RNA polymerases has been shown to play key roles in the initiation, establishment, and maintenance of sRNA-mediated transcriptional silencing (Haag and Pikaard 2011;Nuthikattu et al. 2013;Sasaki et al. 2014).Transcription of the p1 paramutagenic sequences may generate RNA transcripts converted to dsRNA through the RDR activity of MOP1 (Sidorenko and Chandler 2008) and subsequently processed to sRNAs.To test whether paramutagenic sequences are transcribed, we performed run-on transcription assays using nuclei prepared from young sheaths and developing pericarps of active P1-rr of the silenced nontransgenic P1-rr′ and transgenic P1-rr′; P1.2/− genotypes (Fig. 7b and c) and young husks of active P1-rr and silenced nontransgenic P1-rr′ (Fig. 7d).Assays were performed in the presence of 32 P-labeled ribonucleotides, and the resulting 32 P-labeled transcripts were hybridized to RNA probes immobilized on nylon membrane corresponding to the 13, 15α, and 15β subfragments (Fig. 6a; Materials and methods).
The results show that the sequences required for p1 paramutation produced transcripts that hybridize with both forward and reverse probes in all assayed tissues (Fig. 6b-d).The highest transcription levels were observed for the 15β forward and reverse probes, indicating that both DNA strands were strongly transcribed.Strikingly, all tested genotypes had roughly similar transcription levels in all tissues, suggesting that transcription was not reduced in the silenced nontransgenic P1-rr′ and the transgenic P1-rr′; P1.2/− genotypes.This finding contrasts with transcription of the p1 coding sequence which is reduced ∼5X in the silenced P1-rr′ allele relative to the active P1-rr (Sidorenko and Chandler 2008).The results showing similar levels of transcription in the active P1-rr and silenced P1-rr′ regulatory sequences are reminiscent of the results at the maize b1 locus where the sequences required for paramutation were transcribed in the active and transcriptionally silenced b1 alleles at similar levels (Alleman et al. 2006).

α-Amanitin sensitivity implicates RNA Pol-II in transcription of the p1 paramutation sequences
To determine which of the 5 plant DNA-dependent RNA polymerases is responsible for transcription of the p1 paramutagenic sequences, we tested the effects of the fungal toxin α-amanitin, which most strongly inhibits the Pol II polymerase, even at low doses of 5-10 µg/ml (Strain et al. 1971;Haaf and Ward 1996;Haag et al. 2012).Higher doses of α-amanitin (50-100 µg/ml) suppress Pol III activity (Haeusler and Engelke 2006), while Pol I remains largely insensitive to α-amanitin even at high concentrations (De Lange et al. 1985).The remaining 2 plant-specific polymerases involved in RdDM, Pol IV and Pol V, were demonstrated to be insensitive to α-amanitin in vitro (Haag et al. 2012).We assayed the effect of α-amanitin on transcription from the 15β subfragment in young husks of heritably silenced P1-rr′ plants.
Our results revealed a ∼10-and ∼33-fold reduction of transcription within the 15β subfragment in the nuclei treated with 50 and 100 µg/ml of α-amanitin, respectively (Fig. 7e).These levels of transcriptional suppression were similar to those of the Pol II-transcribed maize Ubiquitin2 control (Fig. 7e).As expected, transcription of Pol I-transcribed 18S control was less sensitive to The sRNA sample in lane 2 was from a silenced patterned P1-rr′ that was propagated in the absence of P1.2 transgene for 10 consecutive generations.The sRNA samples loaded in lanes 3 and 4 were from sibling P1-rr′ plants that were homozygous for P1-rr, but differed in P1.2 transgene presence/absence: nontransgenic sibling (lane 3) and transgenic sibling (lane 4).The last 2 lanes contain samples from strongly silenced homozygous P1-rr′ plants that were hemizygous for the 15β and 13 + 15 transgenes.Because sRNA levels in the P1-rr sample were below detection (indicated as BD), the average pixel volumes of P1-rr′ (lane 2) were used to calculate fold change (FC) in lanes 3-6.Average FC relative to P1-rr′ was calculated from 2 technical replicates and is shown below each lane.
α-amanitin; ∼2-fold reduction was observed in the presence of these levels of α-amanitin.Based on the high α-amanitin sensitivity of the 15β transcripts, we conclude that Pol-II RNA polymerase is most likely responsible for transcription of the p1 paramutagenic sequences.
sRNAs are more abundant in P1-rr′ than P1-rr Genetic tests implicated the involvement of the RNA-mediated DNA methylation pathway in p1 paramutation (Sidorenko and Chandler 2008;Sidorenko et al. 2009).To further investigate the role of this pathway in p1 paramutation, we assayed accumulation of sRNAs derived from the p1 paramutagenic sequences.Initial experiments failed to detect sRNA in either P1-rr or P1-rr′ using RNA blots containing 20 µg of total RNA from pericarp or leaf tissues and hybridized with randomly 32 P-labeled 13, 14, and 15 DNA probes (Sidorenko, data not shown).Next, we used a highly sensitive detection method that was developed for assaying low-abundance sRNAs involved in b1 paramutation (Arteaga-Vazquez et al. 2010).Total RNA was extracted from sRNA-rich young ears, and low-molecular-weight RNAs were enriched (Materials and methods) and loaded in high amounts (100 µg per lane).The resulting blots were hybridized with 32 P end-labeled with highly sensitive DNA/LNA oligos (Fig. 4a; the exact location of the DNA/LNA oligos is shown in Supplementary Fig. 1a).sRNA blots hybridized with the complementary forward (F) and reverse (R) DNA/LNA oligos are shown in Fig. 7f and g. and sRNA signal was not detected in active P1-rr (1st lanes), while varying levels of sRNA were observed in the silenced P1-rr′ samples (2nd and 3rd lanes).Specifically, low levels of sRNA were detected in samples from a P1-rr′ lineage that was maintained in the absence of the inducing transgene for 10 consecutive generations (2nd lanes).Relatively this P1-rr′ lineage, higher sRNA levels (2.3-to 3.9-fold) were observed for P1-rr′ that had segregated away from the P1.2 transgene in the prior generation (3rd lanes).Interestingly, sRNA levels in the sample from transgenic P1-rr′; P1.2/− (4th lanes) were similar to that of a nontransgenic P1-rr′ sibling (3rd lanes).Elevated sRNA levels (2.2-to 4.9-fold) were also detected in samples from the silenced P1-rr′ plants containing the 15β and 13 + 15 transgenes (last 2 lanes).In summary, using large quantities of enriched sRNA and highly sensitive DNA/LNA probes enabled detection of increased sRNA levels in nontransgenic and transgenic P1-rr′ samples relative to the active P1-rr for which sRNAs were not detected.

Discussion
Minimal p1 paramutagenic sequence is ∼600 bp and overlaps with a P1-rr enhancer A previous study had shown that a 1.2-kb p1 noncoding fragment (P1.2) could induce paramutation of P1-rr (Sidorenko and Peterson 2001;Sidorenko and Chandler 2008).Here, we used transgenic maize plants to identify the minimal sequences within the P1.2 fragment required for p1 paramutation.The results showed that a 409-bp fragment (15) was sufficient to induce frequent and strong P1-rr silencing.However, the resulting P1-rr-silenced states were less heritable than those induced by the full-length P1.2 fragment.Heritability and secondary paramutation were more frequent in plants containing 15 plus an additional 202 bp of 13 subfragment (Supplementary Fig. 1b).While it remains unclear how addition of the 13 subfragment improves heritability and secondary paramutagenicity of the P1-rr′ state, one possibility may be that transgenes carrying both 15 and 13 subfragments produce sRNAs from the entire required for paramutation p1 enhancer sequence (Fig. 1b), establishing DNA methylation and repressive chromatin that is more effectively inherited and transmitted to the naïve P1-rr in the absence of the silencing transgene.In contrast, the inclusion of sequences within subfragment 14 did not increase P1-rr silencing, heritability, or secondary paramutation of P1-rr silenced states.Thus, our results identify a minimal ∼600-bp sequence within P1.2 sufficient to induce p1 paramutation that is heritable and can mediate secondary paramutation at high frequencies.
Prior studies of P1-rr gene structure revealed that the upstream P1.2 fragment has properties of a transcriptional enhancer (Sidorenko et al. 2000;Sidorenko and Peterson 2001).Analysis of related P1-rw alleles identified a 386-bp sequence overlapping P1.2 (Fig. 1a) that acts as an enhancer of p1 expression in cob glumes (Zhang and Peterson 2005;Zhang et al. 2006).In P1-rr, the 386-bp cob glume enhancer sequence overlaps with minimal paramutagenic P1.2 subfragments 13 and 15 (Fig. 1b; Supplementary Fig. 1a).This overlap suggests that the p1 enhancer and paramutation determinants are functionally related.The involvement of enhancer sequences in paramutation is well established for the maize b1 gene, which regulates plant anthocyanin accumulation.An enhancer of b1 expression is located ∼100-kb upstream of the b1 coding sequence; this long-distance enhancer consists of seven 853-bp tandem (hepta) repeats (Stam, Belele, Dorweiler, et al. 2002;Stam, Belele, Ramakrishna, et al. 2002) and is required for both high-level expression and paramutation.Moreover, the b1 hepta repeats mediate high expression of a heterologous reporter gene in transgenic plants and also induce strong paramutation of a susceptible endogenous b1 allele in multiple transgenic events (Belele et al. 2013).Finally, genetic analyses of the maize pl1 gene indicated a similar association between paramutation and a 3′ enhancer of pl1 expression (Erhard et al. 2013;Hollick 2017).Together, the results from p1, b1, and pl1 studies strongly suggest that enhancer sequences mediate both high expression and paramutation of these maize genes.

P1-rr paramutagenic sequences are present in multicopy repeats
The paramutable P1-rr allele and the paramutagenic epialleles P1-rr′ and P1-pr have identical locus structures (Das and Messing 1993;Sidorenko and Peterson 2001) and contain 4 copies of the ∼600-bp paramutagenic sequence within the 1.2-kb direct repeats flanking the p1 coding sequence (Fig. 1a), whereas, a nonexpressing p1-ww allele that is neutral to paramutation has a single truncated copy of the 1.2-kb repeat (Sidorenko and Peterson 2001;Goettel and Messing 2013b).However, other alleles show that the presence of P1.2 repeats is not sufficient to induce paramutation: for example, the spontaneous P1-pr TP isolate is structurally identical to P1-rr but is not paramutagenic (Sekhon et al. 2007;Sekhon and Chopra 2009).Furthermore, the P1-wr alleles contain multiple repeats of sequences highly similar to P1.2 (97%, not shown), but they do not participate in paramutation (Sidorenko and Peterson 2001;Goettel and Messing 2009).These observations suggest that p1 paramutation may require specific epigenetic states of the paramutagenic repeated sequences.
One epigenetic chromatin mark found associated with the P1.2 repeat sequences in the paramutagenic P1-rr′ and P1-pr epialleles is presence of increased nonsymmetric CHH cytosine methylation (Sekhon et al. 2012;Goettel and Messing 2013b), in addition to high CG and CHG methylation.The P1.2 repeats of the P1-pr TP and P1-wr alleles that do not participate in paramutation have significant CG and CHG methylation, but they have low CHH methylation (Sekhon and Chopra 2009;Sekhon et al. 2012).Because CHH methylation is established via sRNA-mediated mechanisms (Matzke and Mosher 2014), this finding is consistent with the model that RdDM pathways are involved in marking the P1.2 repeat sequences as a result of paramutation (Sidorenko and Chandler 2008;Sidorenko et al. 2009).
In addition to p1, a link between sequence repeats and paramutation has been extensively documented at other maize loci (Chandler et al. 2000;Chandler and Stam 2004;Stam 2009;Hovel et al. 2015;Hollick 2017).At the r1 locus, genetic studies demonstrated that alleles with 4 r1 gene copies were strongly paramutagenic, while alleles with fewer gene copies exhibited reduced paramutagenicity; alleles with a single r1 gene copy were not paramutagenic (Kermicle et al. 1995;Panavas et al. 1999).At the b1 locus, analysis of recombinant alleles differing in number of 853-bp enhancer repeats showed that paramutagenicity was directly proportional to repeat copy number (Stam, Belele, Dorweiler, et al. 2002).These results from the maize r1 and b1 loci strongly link repeat number with strength of paramutation, while the involvement of repeat sequences has so far not been reported for maize pl1 paramutation (Hollick 2017).

fMULE sequences are not likely to play a significant role in transgene-induced p1 paramutation
The P1-rr gene locus harbors multiple transposable elements within its flanking 5.2-kb direct repeats (Fig. 1a), the proximal Pb promoter sequences and introns (Goettel and Messing 2010).Some of these elements, specifically fMULE that are present within the distal P1.2 enhancer and proximal promoter fragment, were proposed to play regulatory roles in spontaneous silencing and paramutation of the endogenous P1-rr allele (Goettel and Messing 2013a).In the proposed model (Goettel and Messing 2013a), sRNA-mediated silencing of MULEs at other genomic locations affects fMULEs within the P1-rr regulatory sequences in trans by a homology-based sRNA-mediated silencing mechanism.Silencing of fMULEs within the P1-rr allele may then lead to spreading of silenced chromatin to the nearby P1-rr enhancer, resulting in establishment, and later maintenance, of heritable transcriptional P1-rr silencing (Goettel and Messing 2013b).Our experiments, however, did not reveal a significant role for MULE sequences in transgene-induced P1-rr paramutation.Specifically, subfragment 14, containing an fMULE sequence (Goettel and Messing 2013a), was neither required nor sufficient for strong P1-rr silencing and paramutation.Similarly, fragment Pb containing a 145-bp MULE sequence (not shown) induced only phenotypically weak P1-rr silencing.Rather, our experiments indicate a critical role for the enhancer sequences located within the P1-rr locus subfragments 15 and 13.These sequences do not contain significant similarities to TEs nor to the enhancer sequences involved in b1 paramutation (Sidorenko, not shown).

Sequences required for p1 paramutation are transcribed by Pol II RNA polymerase
Nuclear run-on assays revealed that the paramutagenic sequences are transcribed in all tested tissues (young sheath, husks, and pericarp) and in all tested genotypes (active P1-rr and silenced P1-rr′ with and without the P1.2 transgene).Transcription is strongly inhibited by moderate doses of α-amanitin, indicating that RNA Pol II is most likely responsible.Although transcripts hybridizing to forward and reverse probes were detected, it remains unclear whether Pol II transcription is bidirectional or if Pol II produces transcripts from 1 strand that serve as a template for RDR-mediated transcription to generate complementary RNA.These results at p1 are similar to those at b1, where α-amanitinsensitive transcription from both strands of the b1 enhancer was detected in samples from both paramutagenic and paramutable alleles (Alleman et al. 2006;Arteaga-Vazquez et al. 2010).
At present, it is not clear how Pol II transcription may regulate paramutation.One possibility is that Pol II-derived transcripts are converted by RDR to dsRNA, which can be processed to sRNAs that can target cytosine methylation and repressive chromatin modifications to the corresponding sequences.In Arabidopsis, expression-dependent silencing of active exogenous transposons exemplifies a functional link between Pol II-derived TE transcripts and RDR6-dependent sRNA production and DNA methylation (Nuthikattu et al. 2013;Fultz et al. 2015;McCue et al. 2015;Panda et al. 2016;Fultz and Slotkin 2017).However, because there is no published evidence that suggests involvement of the maize RDR6 orthologs in paramutation, it remains unclear whether a similar mechanism may act at the P1-rr enhancer sequences.Another possibility is that transcription rather than transcripts may play a role in regulation of enhancer activity and paramutation.A precedent for a role of transcription in gene regulation was described for a human erythroid cell line where expression of the endogenous Cdkn1b gene depended on transcription per se, not transcripts from a downstream long noncoding RNA (Paralkar et al. 2016).Further investigation will be required to discern the role of transcripts and transcription in the mechanism of p1 paramutation.

P1-rr paramutation is associated with increased levels of sRNAs and cytosine methylation of the p1 paramutagenic sequences
Previous genetic results indicated that sRNA-mediated RdDM-like mechanisms are involved in the regulation of p1 paramutation (Sidorenko and Chandler 2008;Sidorenko et al. 2009).Here we confirmed the presence of sRNAs homologous to the paramutagenic p1 enhancer sequences and showed that these sRNAs accumulate to higher levels in silenced P1-rr′ than in P1-rr.Interestingly, sRNAs corresponding to the p1 enhancer repeats are in low abundance, and their detection by Northern blot required loading high amounts of sRNA and probing with sensitive DNA/LNA oligos.The low abundance of sRNAs is similar to that reported for b1 paramutation, where sRNAs corresponding to the b1 enhancer repeats were also in low abundance (Arteaga-Vazquez et al. 2010).However, in contrast to b1 paramutation where similar sRNA levels were detected for paramutagenic and paramutable genotypes (Arteaga-Vazquez et al. 2010), in our experiments, silenced P1-rr′ exhibited higher sRNA levels than the active P1-rr, for which sRNAs were below detection.Notably, sRNA levels were further increased (2.2-to 4.9-fold) in transgenic P1.2, 15β, and 13 + 15 P1-rr′ plants.Similar increases of sRNA levels (∼2-to 3-fold) were also detected for b1 transgenic plants carrying hepta repeats of paramutagenic sequence (Belele et al. 2013).However, at b1, levels of sRNA were reduced in nontransgenic siblings, when we did not detect a reduction of sRNAs in a nontransgenic sample.While additional studies are needed to better understand reasons for the observed differences between p1 and b1 paramutation, together, these genetic and molecular results provide additional support for models of sRNA-mediated mechanisms of maize paramutation (Arteaga-Vazquez and Chandler 2010;Hollick 2017).
DNA blot analyses of silenced P1-rr′ plants revealed increased cytosine methylation of the SalI and PstI restriction sites relative to P1-rr, which is consistent with prior findings (Das and Messing 1994;Sidorenko and Peterson 2001;Sekhon et al. 2012;Wang et al. 2017).The new insights obtained in this study revealed that the extent of methylation of endogenous P1-rr′ largely depended on sequences present within the silencing inducing transgene.Specifically, P1.2 and 13 + 15 that induce highly heritable P1-rr′ fully encompass the assayed endogenous P1-rr SalI and PstI restriction sites, and these sites were fully methylated in P1-rr′.At the same time, Δ14 + 15α, ΔΔ14 + 15, and 15β that induce weakly heritable P1-rr′ have many of the assayed sites located at the edges of sequence identity, and these sites were variably methylated.Reduced methylation of the restriction sites located at the edges of sequence identity between P1-rr and transgenes is not surprising and is similar to the results of prior studies (Sekhon et al. 2012) that showed reduced cytosine methylation of the silenced endogenous P1-rr′ enhancer near the borders of homology to a full-length transgenic P1.2 fragment that was used to induce paramutation.Accounting for the demonstrated role of RdDM in p1 paramutation (Sidorenko and Chandler 2008;Sidorenko et al. 2009), these results are consistent with the hypothesis that sRNAs produced from transgenic P1.2 subfragments drive increased DNA methylation of homologous P1-rr enhancer sequences resulting in enhancer inactivation and transcriptional P1-rr silencing (Sidorenko and Chandler 2008).The involvement of DNA methylation in transgene-induced P1-rr paramutation is similar to its proposed role in paramutation at the b1 and r1 loci.Specifically, at the b1 locus, both endogenous and transgene-induced paramutation correlates with increased cytosine methylation within the tandem hepta repeats required for paramutation (Haring et al. 2010;Belele et al. 2013).At the r1 locus, paramutation is associated with increased cytosine methylation within specific regulatory sequences of a paramutable r1 gene (Walker 1998;Panavas et al. 1999;Walker and Panavas 2001).Together, results from r1, b1, and p1 support involvement of DNA methylation of enhancer sequences in regulation of endogenous and transgene-induced paramutation in maize, although a causal relationship has not been established.

Summary and outlook
This study in transgenic plants identifies a minimal ∼600-bp sequence required to induce p1 paramutation.This sequence occurs in multiple direct repeats near the p1 gene, overlaps with a previously identified P1-rr cob glume enhancer, and is likely transcribed by Pol II RNA polymerase.Transgene-induced paramutation of P1-rr is correlated with increased levels of sRNAs and cytosine methylation of the endogenous P1-rr′ corresponding to the paramutation-inducing sequence.We conclude that this sequence plays important roles in both paramutation and P1-rr transcriptional regulation.The identification of the minimal p1 paramutation sequence will facilitate further biochemical and genetic experiments to identify trans-acting factors mediating P1-rr enhancer function and paramutation.

Fig. 2 .
Fig. 2. Silencing phenotypes of P1-rr induced by the P1.2 deletion transgenes.a) The active P1-rr allele with dark red pericarp and glumes.Photos b through j) show phenotypes of silenced ears.b and c) Weakly silenced patterned ears with slight reduction of pericarp pigmentation and red cob glumes.Enlargements show slightly darker striping on kernel pericarps.d, e) Strongly silenced patterned ears with pigmentation that varies from heavy striped in d to nearly colorless pericarp with residual pigment at the silk scar in e.The color of cob glumes in these ears varies from pink to near colorless.f and g) Weakly silenced blushed ears have light red to orange pericarp with no visible striping.Please, note that cob glume pigmentation is much lighter than in weakly silenced patterned ears shown in b and c. h, i) Strongly silenced blushed ears have pigmented pericarp gown (sides of a kernel) and lighter pigmented or colorless pericarp crown (top of a kernel).The cob glumes have very little pigment and vary from light pink to colorless.j) The strongest transgene-induced silencing phenotype had colorless ears with no visible pigment in pericarp or cob glumes.k, l) Pigment deposition in representative patterned and blushed kernels, respectively.Photo inserts next to each kernel show thin sections of pericarp prepared with handheld razor blade and examined under binocular microscope.In the patterned kernel, cutting through the pigmented stripe showed that pigment accumulates in the crown and outer cell layer of pericarp gown.In the blushed kernel, pericarp crown is colorless while inner layers of pericarp gown are pigmented.Black bars denote the scale in millimeters (mm).Letter "a" indicates aleurone.

Fig. 3 .
Fig. 3. Induction of P1-rr silencing by transgenes carrying P1.2 subfragments.A simplified diagram of P1.2 subfragments is shown on the left of the figure.Location of p1 cob glume enhancer, fMULE, and major restriction sites is shown on top.Shading within transgenic P1.2 subfragments is similar to Fig. 2; black rectangles indicate sequence of the p1 cob glume enhancer, light gray rectangles indicate diagram fMULE, and open rectangles show DNA that is unique to p1 locus.The table in the middle of the figure presents a summary of the number of events, number of plants assayed, and average frequency of P1-rr silencing within the informative transgenic p1-ww/P1-rr*, TR/− plants (Supplementary Fig.2).The average frequency of silencing for a construct was calculated as sum of frequencies of silencing for each event divided by the number of tested events (Materials and methods).Statistically significant differences relative the P1.2 control are indicated by 2 stars for P < 0.001 and 3 stars for P≤0.0001.P-values were calculated by 2-sample test for equality of proportions with Yates' correction for continuity, followed by Bonferroni-Holm adjustment for multiple comparisons (Materials and methods).The box plot on the right shows silencing frequencies for individual events with each of the dots sized proportionally to the number of transgenic plants assayed for that event.Light-and dark-shaded areas of the box plot indicate 2 middle quartiles with the dividing line indicating median frequency of silencing.The whiskers indicate 1.5 times the minimum and maximum of the respective interquartile range.Additional details regarding transgene segregation, results of molecular analyses, and frequencies of strong and weak P1-rr silencing are in Supplementary Fig.3.

Fig. 4 .
Fig. 4. Heritability of P1-rr silencing induced by transgenes carrying P1.2 subfragments.A simplified diagram of P1.2 subfragments is shown on the left of the figure.Location of p1 cob glume enhancer, fMULE, and major restriction sites is shown on top.Shading within transgenic P1.2 subfragments is similar to Fig. 1; black rectangles indicate sequence of the p1 cob glume enhancer, light gray rectangles diagram fMULE, and open rectangles show DNA that is unique to p1 locus.The table in the middle of the figure presents a construct level summary for the informative nontransgenic P1-rr*/P1-ww, −/− plants (Supplementary Fig. 4).Number of events/plants tested and average frequency of silencing (Materials and methods) for each construct are shown.Three stars indicate statistically significant differences (P < 0.0001) relative the P1.2 control.P-values were calculated by 2-sample test for equality of proportions with Yates' correction for continuity, followed by Bonferroni-Holm adjustment for multiple comparisons (Materials and methods).The box plot on the right shows silencing frequencies for individual events with each of the dots sized proportionally to the number of transgenic plants assayed for that event.Light-and dark-shaded areas of the box plot indicate 2 middle quartiles with the dividing line indicating median frequency of silencing.The whiskers indicate 1.5 times the minimum and maximum of the respective interquartile range.Additional information regarding segregation of herbicide tolerance, frequencies of strong, and weak transgene silencing is shown in Supplementary Fig. 5.

5.
Secondary paramutation of silenced P1-rr states induced by transgenes carrying P1.2 subfragments.A simplified diagram of P1.2 subfragments is shown on the left of the figure.Location of p1 cob glume enhancer, fMULE, and major restriction sites is shown on top.Shading within transgenic P1.2 subfragments is similar to Fig. 1; black rectangles indicate sequence of the p1 cob glume enhancer, light gray rectangles indicate diagram fMULE, and open rectangles show DNA that is unique to p1 locus.The table in the middle of the figure presents a construct level summary for the informative nontransgenic P1-rr*/P1-rr, −/− plants (Supplementary Fig. 7).Number of events and plants tested and average frequency of silencing (Materials and methods) for each construct are shown.Three stars indicate statistically significant differences at P < 0.0001, and 1 star indicate significant difference at P < 0.01 relative the P1.2 control.P-values were calculated by 2-sample test for equality of proportions with Yates' correction for continuity, followed by Bonferroni-Holm adjustment for multiple comparisons (Materials and methods).The box plot on the right shows silencing frequencies for individual events with each of the dots sized proportionally to the number of transgenic plants assayed for that event.Light-and dark-shaded areas of the box plot indicate 2 middle quartiles with the dividing line indicating median frequency of silencing.The whiskers indicate 1.5 times the minimum and maximum of the respective interquartile range.Additional information regarding segregation of herbicide tolerance and frequencies of strong and weak transgene silencing is shown in Supplementary Fig. 7.

Fig. 6 .
Fig. 6.Summary of SalI and PstI methylation within transgene-silenced endogenous P1-rr′ allele.DNA blots used to generate this summary are shown in Supplementary Figs.9-11.For both a and b, ovals denote SalI, and diamonds denote PstI restriction sites.EcoRI and KpnI are not sensitive to cytosine methylation, while SalI is sensitive to CpG methylation, and PstI is sensitive to CpNpG methylation.Shading of ovals and diamonds indicates methylation status of the SalI and PstI sites: black, complete methylation; striped, partial methylation; and open, no methylation.The 0.7-kb PstI fragment is shown in gray because this fragment was faint on DNA blots and difficult to reliably observe (Supplementary Figs.9c, 10c, and 11c).a) Map of the endogenous P1-rr gene locus.P1-rr exons (black boxes) and introns (thin lines) (Grotewold et al. 1991) and location of probe 15 (gray boxes) (Lechelt et al. 1989) are shown above the P1-rr map.The P1-rr transcription start site is shown as a bent arrow.The 5.2-kb direct repeats (black rectangles) overlap with the smaller 1.2-kb direct repeats (hatched boxes).Below the P1-rr map, double-headed black arrows indicate the location of the P1.2 transgene fragments.The restriction maps of EcoRI (E) + KpnI (K), SalI, and PstI sites are also shown below the P1-rr map.Double-headed arrows above the SalI and PstI restriction maps indicate fragments recognized by P1-rr probe 15.b) The expanded views of SalI and PstI DNA methylation within the upstream and downstream 1.2-kb direct repeats, as detected by hybridization with probe 15.Methylation of the active endogenous P1-rr allele and transgene-silenced P1-rr′ epiallele is shown on top.The names of the transgenic constructs are shown on the right.Double-headed arrows below the Δ14 + 15α diagram show locations of the missing 2.9-kb band (red) and observed 1.2-kb doublet (black) (Supplementary Fig. 10c).Gray boxes indicate sequence identity of each transgene to the endogenous P1-rr allele.
Fig. 6.Summary of SalI and PstI methylation within transgene-silenced endogenous P1-rr′ allele.DNA blots used to generate this summary are shown in Supplementary Figs.9-11.For both a and b, ovals denote SalI, and diamonds denote PstI restriction sites.EcoRI and KpnI are not sensitive to cytosine methylation, while SalI is sensitive to CpG methylation, and PstI is sensitive to CpNpG methylation.Shading of ovals and diamonds indicates methylation status of the SalI and PstI sites: black, complete methylation; striped, partial methylation; and open, no methylation.The 0.7-kb PstI fragment is shown in gray because this fragment was faint on DNA blots and difficult to reliably observe (Supplementary Figs.9c, 10c, and 11c).a) Map of the endogenous P1-rr gene locus.P1-rr exons (black boxes) and introns (thin lines) (Grotewold et al. 1991) and location of probe 15 (gray boxes) (Lechelt et al. 1989) are shown above the P1-rr map.The P1-rr transcription start site is shown as a bent arrow.The 5.2-kb direct repeats (black rectangles) overlap with the smaller 1.2-kb direct repeats (hatched boxes).Below the P1-rr map, double-headed black arrows indicate the location of the P1.2 transgene fragments.The restriction maps of EcoRI (E) + KpnI (K), SalI, and PstI sites are also shown below the P1-rr map.Double-headed arrows above the SalI and PstI restriction maps indicate fragments recognized by P1-rr probe 15.b) The expanded views of SalI and PstI DNA methylation within the upstream and downstream 1.2-kb direct repeats, as detected by hybridization with probe 15.Methylation of the active endogenous P1-rr allele and transgene-silenced P1-rr′ epiallele is shown on top.The names of the transgenic constructs are shown on the right.Double-headed arrows below the Δ14 + 15α diagram show locations of the missing 2.9-kb band (red) and observed 1.2-kb doublet (black) (Supplementary Fig. 10c).Gray boxes indicate sequence identity of each transgene to the endogenous P1-rr allele.

Fig. 7 .
Fig. 7. Transcription and sRNA accumulation within the sequences required for p1 paramutation.a) Diagram of the P1.2 fragment and the location of probes used in nuclear run-on and sRNA blot analysis.RNA probes used for nuclear run-on analysis are shown as arrows below the P1.2 map.Complementary 45-nt DNA/LNA oligos used for the northern blot analysis of sRNA is shown by a star.The exact location of the 45-bp DNA/LNA probes is diagrammed in Supplementary Fig. 1a.For b through e, the names of RNA probes used for detection of nuclear run-on transcripts are indicated along the x-axis, and the mean signal intensity (pixels/mm 2 ) is indicated on the y-axis.b) Transcription detected with 13, 15α, and 15β probes in nuclei extracted from young sheath.c) Transcription detected 13, 15α, and 15β probes in nuclei extracted from immature pericarps.d) Transcription detected with 13, 15α, and 15β probes in nuclei extracted from young P1-rr and P1-rr′ husks.e) Effect of α-amanitin on transcription detected with the 15β probe in nuclei extracted from young P1-rr′ husks.The percentages of the signal remaining after treatment with 50 µg/ml and 100 µg/ml α-amanitin relative to untreated nuclei are shown above the corresponding bars.f, g) Northern blot analyses of sRNA accumulation in the young ears of active P1-rr, silenced P1-rr′ (no transgene), and transgenic P1-rr′ plants.Forward and reverse DNA/LNA oligos were end-labeled with 32 P and used as hybridization probes (Materials and methods).A photo of the gels stained with ethidium bromide illustrates sRNA loading.Location of 25 and 21 nt of the microRNA marker is shown on the right of each blot.The genotype of each sample is indicated on the top of each lane.The sRNA sample in the first lane was extracted from young P1-rr ears.The sRNA sample in lane 2 was from a silenced patterned P1-rr′ that was propagated in the absence of P1.2 transgene for 10 consecutive generations.The sRNA samples loaded in lanes 3 and 4 were from sibling P1-rr′ plants that were homozygous for P1-rr, but differed in P1.2 transgene presence/absence: nontransgenic sibling (lane 3) and transgenic sibling (lane 4).The last 2 lanes contain samples from strongly silenced homozygous P1-rr′ plants that were hemizygous for the 15β and 13 + 15 transgenes.Because sRNA levels in the P1-rr sample were below detection (indicated as BD), the average pixel volumes of P1-rr′ (lane 2) were used to calculate fold change (FC) in lanes 3-6.Average FC relative to P1-rr′ was calculated from 2 technical replicates and is shown below each lane.