Maelstrom Represses Canonical Polymerase II Transcription within Bi-Directional piRNA Clusters in Drosophila melanogaster

SUMMARY

In Drosophila, 23–30 nt long PIWI-interacting RNAs (piRNAs) direct the protein Piwi to silence germline transposon transcription. Most germline piRNAs derive from dual-strand piRNA clusters, heterochromatic transposon graveyards that are transcribed from both genomic strands. These piRNA sources are marked by the Heterochromatin Protein 1 homolog, Rhino (Rhi), which facilitates their promoter-independent transcription, suppresses splicing, and inhibits transcriptional termination. Here, we report that the protein Maelstrom (Mael) represses canonical, promoter-dependent transcription in dual-strand clusters, allowing Rhi to initiate piRNA precursor transcription. Mael also represses promoter-dependent transcription at sites outside clusters. At some loci, Mael repression requires piRNA pathway, while at others, Mael piRNAs play no role. We propose that by repressing canonical transcription of individual transposon mRNAs, Mael helps Rhi drive non-canonical transcription of piRNA precursors without generating mRNAs encoding transposon proteins.

Graphical Abstract

eTOC Paragraph

Transposon-rich piRNA clusters in flies represent a gene-expression paradox. These genomic loci must be transcribed to produce piRNA precursor transcripts yet transposon promoters must be silenced to prevent transposon derepression. Chang et al. report that Mael helps solve this paradox by specifically repressing canonical, promoter-dependent Pol II transcription within fly piRNA clusters.

INTRODUCTION

In Drosophila melanogaster, 23–30 nt long PIWI-interacting RNAs (piRNAs) direct transposon silencing by serving as guides for Argonaute3 (Ago3), Aubergine (Aub), and Piwi, the three fly PIWI proteins (Aravin et al., 2001; Vagin et al., 2006). In the germ cell cytoplasm, Aub and Ago3 increase the abundance of their guide piRNAs via the ping-pong cycle, an amplification loop in which cycles of piRNA-directed cleavage of sense and antisense transposon-derived long RNAs generate new copies of the original piRNAs in response to transposon transcription (Brennecke et al., 2007; Gunawardane et al., 2007). In addition to amplifying piRNAs, this “ping-pong” pathway also produces long 5′ monophosphorylated RNA that enters the phased piRNA pathway, generating head-to-tail strings of piRNAs bound to Piwi, and, to a lesser extent, Aub (Han et al., 2015; Mohn et al., 2015; Wang et al., 2015). Unlike Ago3 and Aub, Piwi acts in both the germline and the adjacent somatic follicle cells to repress transposon transcription rather than to cleave their transcripts (Li et al., 2009a; Malone et al., 2009). Nuclear Piwi is believed to bind nascent RNA transcripts, and through the protein Panoramix tether the histone methyltransferase SETDB1 to transposon-containing loci. SETDB1, in turn, trimethylates histone H3 on lysine 9 (H3K9me3), a modification required to create repressive heterochromatin (Sienski et al., 2012; Muerdter et al., 2013; Klenov et al., 2014; Sienski et al., 2015; Yu et al., 2015).

piRNA precursor RNAs are transcribed from piRNA clusters, heterochromatic loci that comprise transposons and transposon fragments, thereby recording a species’ evolutionary history of transposon invasion (Brennecke et al., 2007). Drosophila piRNA clusters can be uni-strand, transcribed from one genomic strand, or dual-strand, transcribed from both genomic strands. Uni-strand clusters, such as the ~180 kbp flamenco (flam) locus, silence transposons in somatic follicle cells (Sarot et al., 2004; Mevel-Ninio et al., 2007), whereas dual-strand clusters, such as the ~250 kbp 42AB locus, predominate in the germline (Malone et al., 2009). Some uni-strand clusters, such as cluster 2, are active in both tissues.

Canonical, promoter-initiated, RNA polymerase II (Pol II) transcription generates spliced, polyadenylated precursor piRNAs from flam (Mevel-Ninio et al., 2007; Goriaux et al., 2014). In contrast, most dual-strand clusters do not use standard promoters. Instead, the Heterochromatin Protein 1 paralog Rhino (Rhi) binds to H3K9me3 present on the piRNA cluster chromatin, and in conjunction with its protein partners, drives non-canonical transcription (Volpe et al., 2001; Klattenhoff et al., 2009; Zhang et al., 2012; Le Thomas et al., 2014; Mohn et al., 2014; Zhang et al., 2014). One Rhi-associated protein, Moonshiner (Moon), is a germline-specific TFIIA-L paralog that allows Pol II to initiate transcription without promoter sequences, allowing every bound Rhi to be a site of potential transcriptional initiation (Andersen et al., 2017). Another Rhi-binding protein, Cutoff (Cuff), suppresses splicing and transcriptional termination (Pane et al., 2011; Mohn et al., 2014; Zhang et al., 2014; Chen et al., 2016). Thus, Rhi promotes non-canonical transcription: RNA synthesis initiating at many sites throughout both strands of a dual-strand cluster, in contrast to the canonical, promoter-dependent transcription of flam and conventional protein-coding genes.

Maelstrom (Mael), a protein with HMG- (Findley, 2003) and MAEL- (Zhang et al., 2008a) domains, has been suggested to play multiple roles in Drosophila oogenesis and mouse spermatogenesis, including repression of the microRNA miR-7 (Pek et al., 2009), transposon silencing (Lim and Kai, 2007; Soper et al., 2008; Sienski et al., 2012), heterochromatin formation (Pek et al., 2009; Sienski et al., 2012), and piRNA production (Castaneda et al., 2014). Here, we report that Mael suppresses canonical transcription both within and outside dual-strand piRNA clusters. In mael mutant ovaries, piRNA cluster heterochromatin organization is largely unaltered, but transcription initiates from previously silent canonical Pol II promoters, including sites within dual-strand clusters. Transcriptional repression mediated by Mael occurs at many sites across the genome; at some, Mael collaborates with the piRNA pathway, while at others Mael-repression is piRNA-independent. We propose that Mael represses promoter-driven transcription of individual, potentially active transposons, allowing Rhi to transcribe such transposon sequences into intron-containing piRNA precursors with little potential to be translated into proteins required for transposition.

RESULTS

Maelstrom Represses Canonical Transcription in Dual-Strand Clusters

Without Mael, both somatic and germline transposons produce long RNA transcripts (Sienski et al., 2012; Muerdter et al., 2013; Pek et al., 2009), while protein-coding genes are largely unaffected (Figure S1). For example, in the germline of mael^M391/r20 ovaries, steady-state RNA abundance from telomeric transposons increased >360-fold for HeT-A and TART and ~49-fold for TAHRE (n = 3; Figure S1). Intriguingly, RNA increased >13-fold from two individual gypsy12 long terminal repeat (LTR) transposon insertions: one in the dual-strand piRNA cluster 42AB (at 42A14; hereafter gypsy12^42AB) and one in the dual-strand piRNA cluster cluster62 (at 40F7; hereafter gypsy12^cluster62; Figures 1 and S1). The same two gypsy12 elements are also desilenced in Rhi- or Cuff-deficient ovaries (Zhang et al., 2014; Mohn et al., 2014). As in rhi and cuff, but unlike wild-type, RNA from the two gypsy12 LTRs was spliced in mael^M391/r20 mutant ovaries. The increase in steady-state gypsy12 LTR RNA from these two loci in mael^M391/r20 mutants reflects a concomitant increase in nascent transcription as measured by global run-on sequencing (GRO-seq; Figure 1; Core et al., 2008). Lysine 4 trimethylation of histone H3 (H3K4me3), a chromatin mark associated with active, promoter-driven transcription (Bernstein et al., 2002; Santos-Rosa et al., 2002; Schneider et al., 2004), also increased at both gypsy12^42AB and gypsy12^cluster62 (>3-fold and >9-fold, respectively, n = 2; Figure 1). These data suggest that in the absence of Mael, RNA polymerase II initiates canonical transcription within the gypsy12 LTR.

Figure 1. Canonical Transcription Initiation in Clusters 42AB, 62, and 38C1 Without Mael.

RNA abundance (RNA-seq), nascent RNA abundance (GRO-seq), and protein density for H3K4me3, H3K9me3, Rhi, and HP1a (ChIP-seq) at the dual-strand piRNA clusters 42AB, cluster62, and 38C1 was measured for wild-type control and mael^M391/r20. Mutant ovaries are shown in red while wild-type controls are shown in black. Annotated transcription start (green) and termination sites (red) are labeled. Data are reads mapping uniquely to the genome from a representative experiment.

See also Figure S1.

A combination of canonical and Rhi-dependent transcription has been proposed to produce piRNA precursor RNA from dual-strand cluster 38C1: two promoters flanking the cluster initiate canonical transcription, while Rhi ensures non-canonical, promoter-independent transcription within the cluster (Mohn et al., 2014; Andersen et al., 2017). Loss of Mael led to increased use of the two canonical cluster 38C1 promoters: in mael^M391/r20 ovaries, transcription initiating at the TATA-box sequences flanking the cluster increased for both plus (mean increase mael/control = 3 ± 1; p = 0.046) and minus (mean increase mael/control = 5 ± 2; p = 0.014) genomic strands (Figure 1). Similarly, the steady-state abundance of cluster 38C1 RNA >150 nt long increased −15-fold in mael^M391/r20 ovaries (mean mael/control = 15 ± 5; p = 3.1 χ 10⁻⁴). However, the density of the active chromatin mark H3K4me3 at the flanking promoters of cluster 38C1 was essentially unchanged in mael^M391/r20 ovaries (mean mael/control = 1.2 ± 0.2; p = 0.34 for the left promoter; mean mael/control = 1.4 ± 0.4; p = 0.26 for the right promoter; Figure 1). We speculate that the absence of a change in H3K4me3 in mael^M391/r20 mutant ovaries reflects the pre-existing, canonical, promoter-driven transcription at cluster 38C1 and its accompanying high levels of H3K4me3.

Together, these data suggest that Mael is required for repression of both transposons outside piRNA clusters and canonical transcription within dual-strand piRNA clusters.

A Reporter for Canonical Transcription in Dual-Strand Clusters

piRNA clusters are highly repetitive, complicating bioinformatic analyses. To test the idea that Mael represses canonical transcription within dual-strand clusters, we used a fly strain (P{GSV6}42A18) bearing a GAL4-responsive gfp transgene inserted into cluster 42AB. The transgene contains five tandem UAS repeats and a core promoter derived from Hsp70Bb (Figure S2A; Toba et al., 1999). In the presence of the germline-specific transcriptional activator nanos-Gal4, the non-repetitive gfp sequence provides a proxy for canonical euchromatic transcription within piRNA clusters. Both gfp mRNA and protein were undetectable even in the presence of nanos-GAL4-VP16 (Figures 2A and 2B). Like 42AB itself, the gfp transgene has a high density of H3K9me3, HP1a, and Rhi across its sequence. Moreover, the reporter construct produces more piRNAs from the (+) genomic strand, as is true for the region of 42AB into which it is inserted. Finally, production of sense and antisense piRNAs from the gfp transgene requires Rhi, Cuff, Piwi, and Armi, proteins all required for piRNA production from dual-strand clusters (Figure S2B; Han et al., 2015).

Figure 2. Mael Represses the Canonical Transcription of a Euchromatic Reporter in 42AB.

(A) RNA-seq, GRO-seq, piRNA, and H3K4me3, H3K9me3, Rhi, and HP1a ChIP-seq profiles for P{GSV6}42A18, a gfp transgene inserted in 42AB containing five tandem GAL4-binding upstream activating sequences with a core promoter derived from the Hsp70Bb gene (Toba et al., 1999). The gfp transgene contains an intron and canonical poly(A) sites in the 3′ UTR. Reads from mael^M391/r20 mutants are shown in red while wild-type controls are shown in black. Annotated transcription termination sites (red) are labeled. RNA- and GRO- seq data are for uniquely mapping reads, while piRNA data are from both uniquely and all mapping reads from the mean of three biological samples. ChIP-seq data are for uniquely mapping reads from the mean of two biological samples.

(B) Representative Western blots for GFP, Mael, or α-Tubulin (α-Tub) from ovaries with the genotype given below. GFP Western signal is the mean of three biological replicates normalized to α-Tub and is given in arbitrary units. p-values were measured using an unpaired, two-tailed t-test compared to w¹¹¹⁸; P{GSV6}42A18/+; +. Uncropped gel images can be found in Figure S2C.

(C) RNA-seq and piRNA profiles for P{GSV6}zip, which is identical in sequence to P{GSV6}42A18 but inserted into the first intron of the subtelomeric gene zip. RNA-seq data are for uniquely mapping reads, while all mapping piRNAs generated by the ping-pong and phased piRNA pathways are shown.

(D) Strategy to identify putative ping-pong and phased piRNAs, and to measure the probability of distances from the 5′ ends of putative ping-pong piRNAs to the 5′ ends of putative phased piRNAs mapping to P{GSV6}zip. Distance probability analyses are for all sequencing reads ≥24 nt. Red lines show non-parametric regression (LOWESS) of the data.

See also Figures S2–S4.

In mael^M391/r20 mutant ovaries, the P{GSV6}42A18 transgene driven by GAL4-VP16 produced correctly spliced gfp mRNA that terminated at a canonical polyadenylation signal sequence; the appearance of gfp mRNA was accompanied by increased transcription (mean mael/control = 80 ± 60; n = 3; p = 3.0 χ 10⁻³) and H3K4me3 (>3-fold increase) across the gfp transgene (Figure 2A). Moreover, the gfp mRNA in mael^M391r20 mutants was translated into full-length GFP protein (Figures 2B and S2C). Finally, a transgene encoding FLAG-Mael restored repression of gfp in mael^M391/r20, demonstrating that loss of Mael, not a secondary mutation, caused inappropriate GFP expression from the transgene inserted in 42AB (Figure 2B).

The number of piRNAs antisense to gfp mRNA decreased ~13-fold (n = 3, p = 10⁻⁴) in mael^M391/r20 (Figure 2A). In theory, gfp derepression in mael^M391/r20 could be explained by the loss of piRNA-directed silencing. Moreover, the P{GSV6}42A18 transgene contains 248 bp of Hsp70 5′ UTR sequence, which is complementary to endogenous antisense piRNAs from the endogenous Hsp70 locus (DeLuca and Spradling, 2018; Huang et al., 2018). To test the possibility that reporter derepression reflects impaired piRNA-directed silencing, we examined the effect of loss of Mael on gfp expression in a fly strain bearing an identical gfp transgene inserted in the first intron of the heterochromatic, subtelomeric protein-coding gene, zip (P{GSV6}zip; Figure 2C). zip is expressed in the germline, and P{GSV6}zip reflects the expression of zip, producing full-length, spliced mRNA in control ovaries (Figure 2B). Like P{GSV6}42A18, expression of P{GSV6}zip increased without Mael: P{GSV6}zip produced >10-fold more gfp mRNA in mael^M391/r20 ovaries (mean mael/control = 11 ± 2; n = 3; p = 7.9 χ 10⁻⁵). Unlike P{GSV6}42A18, P{GSV6}zip generated more rather than fewer gfp piRNAs without Mael (mean mael/control = 14 ± 3; n = 3; p = 4.4 χ 10⁻⁴). Nearly all of these piRNAs derived from the gfp mRNA; few were from the intron in the gfp 3′ UTR. These data suggest that without Mael, increased transcription of P{GSV6}zip provides more mRNA both for translation into GFP and for processing into piRNAs. We conclude that loss of piRNA-directed silencing alone cannot explain the derepression of both P{GSV6}42A18 and P{GSV6}zip in mael^M391/r20.

The arrangement of piRNAs within the gfp mRNA from P{GSV6}zip suggests that piRNA production is initiated by endogenous antisense Hsp70 piRNAs: the majority of phased, sense gfp piRNAs map immediately downstream of sites predicted to be cleaved by initiator Hsp70 piRNAs in both mael^M391/r20 and control ovaries (Figure 2C). To test whether Hsp70 piRNAs initiate phased piRNA production from spliced P{GSV6}zip transcripts, we measured the distance from the 5′ ends of piRNAs which both have a ping-pong partner on the opposite genomic strand and arise from the Hsp70 region of the reporter to the 5′ ends of piRNAs without a ping-pong partner: i.e., the 5′-to-5′ distance from the putative responder piRNAs to the putative phased piRNAs (Gainetdinov et al., 2018). As expected if Hsp70 piRNAs initiate phased piRNA production from the reporter transcript, the probability plot exhibited periodic peaks corresponding to the 5′ ends of phased piRNAs downstream from initiating Hsp70 ping-pong pairs (Figure 2D). Together, our data demonstrate that gfp de-repression in mael^M391/r20 mutant ovaries does not reflect a loss of piRNAs targeting the gfp reporter, but rather corresponds to an independent function of Mael in repressing canonical transcription.

Many Pol II Promoters Normally Repressed within Dual-Strand Clusters are Activated in mael Mutant Ovaries

Without Mael, RNA accumulates from both individual transposons outside clusters (Sienski et al., 2012; Muerdter et al., 2013) and transposon sequences within heterochromatic piRNA dual-strand clusters (Figures S1 and S3A). To further test the idea that Mael represses canonical transcription at sites of Rhi-driven non-canonical transcription within and outside dual-strand clusters, we examined in more detail changes in the transcription of transposons in mael^M391/r20 ovaries.

Among those individual transposons whose steady-state mRNA level changed significantly (increased or decreased ≥ 2-fold; FDR ≤ 0.05) in mael^M391/r20 ovaries, the overwhelming majority were derepressed: steady-state RNA levels increased for 402 of 410 transposons within piRNA clusters and for 1,068 of 1,075 transposons outside clusters (Figure S3A). Among the derepressed transposons, 180 overlapped H3K4me3 peaks whose area also more than doubled in mael^M391/r20 mutants (69 within and 111 outside clusters; Figure S3A). Moreover, of these 180 transposon loci, spliced transcripts—measured by the abundance of uniquely mapping exon-exon junction RNA-seq reads—more than doubled for 29 (13 within and 16 outside clusters) in mael^M391/r20 ovaries (Figure S3A). To obtain a more global view, we analyzed H3K4me3 occupancy around the 5′ ends of the transposons whose increase in steady-state mRNA levels was statistically significant (FDR ≤ 0.05). Compared to control ovaries, the mean H3K4me3 signal in mael^M391/r20 ovaries did not change for protein-coding genes, but increased ~2-fold for the derepressed transposons (Figure S3B).

Consistent with an increase in canonical transcription, we also detected a fourfold increase in nascent transcripts from piRNA clusters (Figures S1 and S4). We conclude that loss of Mael increases canonical Pol II transcription from transposons both outside and within piRNA clusters.

Heterochromatin is Largely Intact in mael Mutant Ovaries

Does loss of heterochromatin at transposons within dual-strand clusters explain the increase in their transcription in mael^M391/r20 ovaries? We examined the density in wild-type and mael^M391/r20 ovaries of H3K9me3, HP1a, and Rhi on the gfp insertion in 42AB, the gypsy12 elements in clusters 42AB and 62, and across cluster 38C1. Without Mael, we observed a modest (1.4–2.8-fold) decrease in the repressive heterochromatin mark H3K9me3 at the gfp insertion in 42AB, gypsy12^42AB, gypsy12^cluster62, and cluster 38C1 (Figures 1 and 2A). HP1a binds H3K9me3, compacts chromatin, and, like Rhi, decorates piRNA clusters (Bannister et al., 2001; Jacobs and Khorasanizadeh, 2002; Nielsen et al., 2002; Vermaak and Malik, 2009; Klenov et al., 2014). We detected only small changes in HP1a or Rhi occupancy of the gfp reporter, gypsy12^42AB, gypsy12^cluster62, or cluster 38C1 (Figures 1 and 2A).

Despite retaining the hallmarks of heterochromatin, gypsy12^42AB, gypsy12^cluster62, and the gfp transgene, all acquired the active transcription mark H3K4me3 when they became transcribed in mael^M391/r20 mutant ovaries (Figures 1 and 2A). In Drosophila, coexistence of active and repressive chromatin marks is typical for active genes embedded in heterochromatin (Riddle et al., 2011). We conclude that failure to repress canonical transcription within piRNA clusters in mael mutants does not reflect a loss of heterochromatin.

Moreover, loss of Mael had no detectable effect on heterochromatin elsewhere in the genome. For the vast majority of transposon families and piRNA clusters, the density of H3K9me3, HP1a, and Rhi differed by less than twofold between mael^M391/r20 and control ovaries (Figure 3). We conclude that Mael is not required to maintain H3K9me3 on chromatin in the germline. Similar results have been reported for Mael in the follicle cell-like, cultured ovarian somatic cell line OSC (Sienski et al., 2012).

Figure 3. Loss of Mael does not alter Heterochromatin.

(A) ChIP-seq was used to measure the change between mael^M391/r20 and control ovaries in the density of H3K9me3, HP1a, or Rhi for transposons between mael^M391/r20 or w¹¹¹⁸ control ovaries. Hashed grey line, ≥2-fold change. Transposons were classified according to Wang et al. (Wang et al., 2015). Data are for uniquely mapping reads from the mean of two biological samples.

(B) ChIP-seq was used to measure the change between mael^M391/r20 and control ovaries in the density of H3K9me3, HP1a, or Rhi for piRNA clusters between mael^M391/r20 or w¹¹¹⁸ control ovaries. Hashed grey line, ≥2-fold change. Data are for uniquely mapping reads from the mean of two biological samples.

(C) ChIP-seq was used to measure the change between mael^M391/r20 and control ovaries in the density of H3K9me3, HP1a, or Rhi for piRNA clusters, transposons, and protein-coding genes. Circle denotes the median. Left and right whiskers encompass the first to third quartiles. Density was calculated using a 1 kbp sliding window with a 500 bp step and was normalized to the total number of reads mapping to each chromosome arm. Data are for uniquely mapping reads from two biological samples.

mael Mutants Produce Fewer piRNAs from Dual-Strand Clusters

Although nascent and steady-state RNA abundance from piRNA clusters increased in mael^M391/r20 mutant ovaries, piRNA abundance decreased (all piRNAs normalized to total miRNAs; mael^M391/r20/control = 0.6 ± 0.2, n = 9). The abundance of piRNAs from gypsy12^42AB (mael/control = 0.060 ± 0.002, n = 3; p = 7.0 χ 10⁻⁹), gypsy12^cluster62 (mael/control = 0.030 ± 0.003, n = 3; p = 4.1 χ 10⁻⁶), the gfp transgene inserted in 42AB (mael/control = 0.09 ± 0.01, n = 3; p = 3.56 χ 10⁻⁷), as well as from cluster 38C1 (mael/control = 0.10 ± 0.01, n = 3; p = 5.92 χ 10⁻⁷), were all more than 10 times lower (Figures 2A and 4A). The loss of piRNAs was particularly acute for some dual-strand piRNA clusters. For example, piRNA abundance decreased 3–5-fold for the paradigmatic uni-strand clusters cluster2 (mael/control = 0.21 ± 0.02; p = 8.4 χ 10⁻⁶) and flam (mael/control = 0.33 ± 0.03; p = 1.8 χ 10⁻⁵), but 32-fold for the dual-strand cluster 42AB (mael/control = 0.053 ± 0.005; p = 9.7 χ 10⁻¹⁰) and 7.6-fold for 80F (mael/control = 0.13 ± 0.01; p = 4.6 χ 10⁻⁷). piRNA abundance also declined for telomeric clusters (mael/control = 0.031 ± 0.001; p = 1.0 χ 10⁻⁸). These data suggest that the canonically transcribed RNA produced from dual-strand clusters in the absence of Mael cannot efficiently enter the piRNA biogenesis pathway.

Figure 4. Fewer piRNAs in mael^M391/r20 Mutants.

(A) Scatter plot comparing piRNAs that uniquely map to clusters, and germline, soma, intermediate, or unknown transposons. The grey line signifies a ≥2-fold change. Data are from three biologically independent samples.

(B) Ping-pong analysis for all or uniquely mapping total piRNAs. Data are mean ± S.D. (n = 3).

(C) Phasing analysis for the distances from the 3′ ends of upstream piRNAs to the 5′ ends of downstream piRNAs on the same genomic strand. Data are for uniquely mapping reads.

All data are mean ± S.D. (n = 3).

See also Figure S5.

The decreased abundance of piRNAs from dual-strand clusters in mael^M391/r20 mutants was accompanied by a marked loss of ping-pong amplification of piRNAs from these clusters (Figures 4A and 4B). This loss did not reflect a defect in the ping-pong machinery per se (all piRNAs: mael^M391/r20, Z₁₀ = 59; control, Z₁₀ = 67). Instead, mael^M391/r20 ovaries showed no significant ping-pong among piRNAs unambiguously mapping to dual-strand clusters (within clusters: mael^M391/r20, Z₁₀ = 0.58; control, Z₁₀ = 23). In contrast, piRNAs mapping outside clusters continued to be amplified (outside clusters: mael^M391/r20, Z₁₀ = 8.1; control, Z₁₀ = 19; Figure 4B). The machinery required to produce phased piRNAs also appears intact in mael^M391/r20 mutants. Despite the reduced abundance of piRNAs in mael^M391/r20 ovaries, significant tail-to-head piRNA phasing remained (all piRNAs: mael^M391/r20, Z₀ = 11 versus control, Z₀ = 14; Figure 4C). Unlike ping-pong amplification, cluster-mapping piRNAs continued to be produced by the phased piRNA pathway (within clusters: mael^M391/r20, Z₀ = 9.0 versus control, Z₀ = 12; outside clusters: mael^M391/r20, Z₀ = 14 versus control, Z₀ = 16; Figure 4C). Our data suggest that Mael plays little or no role in the production of phased piRNAs.

Armi and Piwi, but not Rhi, Repress Canonical Transcription in Dual-Strand Clusters

The current model for fly piRNA biogenesis places Armi upstream and Mael downstream of Piwi (Malone et al., 2009; Haase et al., 2010; Saito et al., 2010; Sienski et al., 2012; Czech et al., 2013; Pandey et al., 2017; Rogers et al., 2017). Consistent with this model, loss of either Armi or nuclear Piwi phenocopies loss of Mael. For example, cluster-mapping steady-state transcript abundance increased to similar levels in armi^72.1/G728E, piwi^2/Nt, and mael^M391/r20 ovaries compared to control (Figure S5A). The abundance of RNA from cluster 38C1, gypsy12^42AB, and gypsy12^cluster62 increased in all three mutants (Figure 5A). Moreover, the RNA produced from both the gypsy12^42AB and gypsy12^cluster62 LTRs was spliced, consistent with a failure to repress canonical transcription. Loss of Piwi or Armi similarly increased the steady-state abundance of RNA and protein from the nanos-Gal4-driven gfp transgene inserted in piRNA cluster 42AB (Figures 5B and 5C). Finally, piwi^2/Nt, armi^72.1/G728E, and mael^M391/r20 mutant ovaries all had fewer gfp-mapping piRNAs than control (Figure S2B; Han et al., 2015).

Figure 5. Armi and Piwi, but not Rhi, are Required to Repress Canonical Transcription in Dual-Strand Clusters.

(A) RNA-seq profiles for cluster 38C1 (left), gypsy12^42AB (center), gypsy12^cluster62 (right), and (B) P{GSV6}42A18 from control, armi^72.1,G128E, piwi^2/Nt, rhi^2/KG, mael^M391/r20, and rhi^2KG; mael^M391/r20, and germline-specific rhi(RNAi) and cuff(RNAi) ovaries.

(C) Representative Western blots and quantifications for GFP, Mael, or α-Tubulin (α-Tub) from ovaries. Uncropped gel images can be found in Figure S2C.

As expected from the role of Cuff in promoting non-canonical transcription of dual-strand clusters (Mohn et al., 2014; Zhang et al., 2014; Chen et al., 2016), gfp mRNA from the reporter in cluster 42AB was not detected in ovaries depleted of germline cuff (Figure 5B). Similarly, consistent with the role for Rhi in sustaining dual-strand cluster transcription (Klattenhoff et al., 2009), the 42AB gfp reporter produced neither gfp mRNA nor protein in rhi^2/KG mutants (Figures 5B and 5C). Without Rhi, 42AB likely becomes a conventional heterochromatic locus, and the gfp reporter remains repressed by Mael. Moreover, the reporter does not produce piRNAs in rhi^2/KG mutants (Figure S2B), suggesting that Mael is guided to its target by a piRNA-independent mechanism. In support of this idea, loss of both Mael and Rhi derepressed the 42AB gfp reporter (Figure 5B). Conceptually, derepression of P{GSV6}42A18 in rhi^2/KG; mael^M391/r20 double mutants is similar to P{GSV6}zip in mael^M391/r20 single mutants: both loci are heterochromatic but neither functions as a piRNA cluster (Figures 2C and 5B).

Notably, the gfp reporter in 42AB produced less mRNA and protein in rhi^2/KG; mael^M391/r20 double mutants than in mael^M391/r20 single mutants (Figures 5B and 5C). Why does loss of Rhi reduce gfp reporter expression? Perhaps Rhi creates a chromatin environment that is transcriptionally permissive for euchromatic genes but repressive for heterochromatic genes. We speculated that low levels of Rhi might suffice to support canonical transcription of the gfp transgene, even in the presence of Mael. To test this idea, we used RNAi to reduce, but not eliminate, Rhi in the ovary germline (Figures S5B and S5C). Consistent with the hypothesis, rhi(RNAi) ovaries produced spliced gfp mRNA, unlike rhi^2/KG null mutants (Figure 5B). To test that Rhi was directly responsible for the increase in gfp transcript abundance, we combined rhi(RNAi) with the rhi^2/KG null mutation (rhi^2/KG; rhi(RNAi)). No gfp mRNA was detected in rhi^2/KG; rhi(RNAi) ovaries (Figure S5D).

These results indicate that Rhi paradoxically enables the canonical transcription of gfp. The overall loss of piRNA cluster transcripts between rhi(RNAi) and rhi^2/KG null mutant ovaries was similar, implying that most regions in piRNA clusters are not acutely sensitive to Rhi dosage (Figure S5E). Together, the data suggest that by enabling RNA Pol II transcription in otherwise repressive heterochromatin, Rhi creates the requirement for Mael to suppress canonical transcription in dual-strand piRNA clusters. Therefore, in dual-strand piRNA clusters, Mael serves to repress the canonical transcription enabled by Rhi as a side-effect of its promoting non-canonical transcription.

Previous studies showed that Mael represses individual transposons outside clusters in ovarian somatic cells (Sienski et al., 2012). We asked whether Mael is required in vivo in somatic follicle cells to repress canonical transcription from a euchromatic gene inserted in heterochromatin. Neither 42AB nor rhi is expressed in follicle cells (Volpe et al., 2001; Klattenhoff et al., 2009; Malone et al., 2009), so in this tissue the gfp reporter inserted in 42AB, P{GSV6}42A18, corresponds to a euchromatic, canonical transcription unit present in heterochromatin. We used the follicle cell-specific driver, Tj-GAL4, to promote canonical transcription of P{GSV6}42A18. Without Mael, GFP protein abundance increased >3-fold compared to control (mael/control = 3.3 ± 0.9; n = 3, p = 0.004). These results confirm that in both the germline and the soma, Mael acts to repress canonical transcription in heterochromatin.

Rhi is not required to initiate transcription at the flanking promoters of cluster 38C1, but is necessary for promoter-independent initiation for both strands within the cluster (Mohn et al., 2014; Andersen et al., 2017). Without Rhi, canonical transcription initiating at the flanking promoters terminates at polyadenylation sites ~400 bp from the transcription start sites (Figure 5A; Mohn et al., 2014; Andersen et al., 2017). The role of Mael in repressing canonical transcription predicts that further loss of Mael in rhi^2/KG mutants will increase transcription of cluster 38C1. Indeed, rhi^2/KG; mael^M391/r20 double mutants had more cluster 38C1 transcripts than rhi^2/KG single mutants: rhi^2/KG; mael^M391/r20 was 11.8 ± 5.8 times greater than rhi^2/KG alone (p = 2.7 χ 10⁻⁴; rhi^2/KG/control = 0.3 ± 0.2, p = 10⁻⁴; rhi^2/KG; mael^M391/r20/control = 4.1 ± 0.3, p = 10⁻⁴; Figure 5A). These results suggest that Mael also represses canonical transcription at cluster 38C1.

Both Rhi and Mael are Required to Repress gypsy12 LTR Transcription

Apart from its role in ensuring transcription of dual-strand piRNA clusters, Rhi also appears to have a separate function in repressing several transposons (Klattenhoff et al., 2009). For example, without Rhi gypsy12^42AB and gypsy12^cluster62 LTRs within dual-strand clusters are desilenced (Zhang et al., 2014; Mohn et al., 2014). In the absence of Mael, the two gypsy12 LTRs are also derepressed. Do Rhi and Mael function additively to repress gypsy12? Consistent with the two proteins functioning non-additively, steady-state mRNA levels of both gypsy12 LTRs do not further increase when both Mael and Rhi are removed compared to the single mutants (Figure 5A). Future studies to unravel the mechanisms of transcription initiation within heterochromatin should help elucidate how Rhi and Mael contribute to silencing.

Mael Represses Transposons by both piRNA-dependent and piRNA-independent mechanisms

In mael^M391/r20 mutants, piRNAs mapping outside clusters continue to be amplified by ping-pong (Figure 4B). However, the plurality of these ping-pong pairs were derived from individual R1 non-LTR retrotransposons. In mael^M391/r20 mutants, even though R1 piRNAs compose just ~8% of all piRNAs uniquely mapping outside clusters, they correspond to ~26% of all ping-pong piRNAs mapping outside clusters Curiously, R1 piRNA abundance increased ~4-fold, and both antisense and sense R1 transcript abundance increased ~9-fold in mael^M391/r20 mutants. This concomitant increase in R1 RNA and piRNAs suggests that, like the gfp reporter in the intron of the subtelomeric zip gene, (1) loss of Mael causes increased canonical transcription from individual R1 promoters, and (2) that at sites outside piRNA clusters, the increased RNA fuels increased piRNA production. These data provide further evidence that Mael can function to repress canonical transcription by a mechanism independent of piRNA production.

To test whether Mael represses other genomic loci by a piRNA-independent mechanism, we asked if the same transposons are derepressed in mael^M391/r20 and piwi^2/Nt mutants. Among individual transposons derepressed in mael^M391/r20, steady-state mRNA levels were higher in mael^M391/r20 than in piwi^2/Nt mutants for 110 of 402 transposons within clusters and for 324 of 1,068 transposons outside clusters (≥2-fold difference; FDR ≤0.05; Figure 6A). Thus, Mael represses ~30% of individual transposon loci—within and outside clusters—by a piRNA-independent mechanism. Moreover, the H3K4me3 occupancy near the 5′ ends of these piRNA-independent Mael-repressed transposons increased ~2-fold (Figure 6B). Only ~3.5% of transposons had higher steady-state mRNA levels in piwi^2/Nt than in mael^M391/r20 (13 of 402 transposons within clusters, and 42 of 1,068 transposons outside clusters; ≥ 2-fold difference; FDR ≤ 0.05; Figure 6A). Together, these data suggest that repression by Mael can be guided by piRNAs or act independent of the piRNA pathway.

Figure 6. Mael represses transposons by piRNA-dependent and piRNA-independent mechanisms.

(A) Scatter plots comparing sense steady-state RNA (RNA-seq) abundance between mael^M391/r20 and piwi^2/Nt ovaries. All data are the mean of six independent biological samples for uniquely mapping reads.

(B) H3K4me3 ChIP-seq data from 5 kb up and downstream of the 5′ ends of the individual transposon insertions with increased steady-state mRNA levels in mael^M391/r20 compared to piwi^2/Nt ovaries (110 within clusters and 324 outside clusters from Figure 6A). Individual control and mael^M391/r20 replicates are shown and mean H3K4me3 signal surrounding the 5′ ends is shown above, while signal from individual transposon insertions is show below.

Maelstrom is a Suppressor of Position-Effect Variegation

Position-effect variegation (PEV) occurs when a phenotype varies from cell to cell or fly to fly because heterochromatin can spread across the causative gene by chance (Muller, 1930; Elgin and Reuter, 2013). Suppressors of PEV are often genes involved in making or maintaining heterochromatin, such as Suppressor of variegation 205 (Su(var)2-5), which encodes HP1a (Eissenberg et al., 1992). The loss of gfp silencing in mael^M391/r20 mutants is reminiscent of a loss-of-function mutation in a suppressor of PEV (Figures 2A and 2C). In addition to UAS-driven gfp, the P{GSV6}42A18 transgene contains the mini-white gene (w^+mc). Like gfp, w^+mc was silenced in control flies: w¹¹¹⁸; P{GSV6}42A18 flies had white eyes. In contrast, w^+mc was active in mael mutants: the eyes of w¹¹¹⁸; P{GSV6}42A18; mael^M391/r20 flies were orange (Figure S6).

To test if mael is a suppressor of PEV in other genomic contexts, we employed two additional transgenic PEV models: P{EPgy2}EY08366, which inserts w^+mc near 42AB, and P{EPgy2}DIP16^EY02625, which inserts w^+mc near flam. PEV at both reporters has been shown to be suppressed by Su(var)2-5 or Su(var)3-9 mutations, but not by piwi or aub mutations (Moshkovich and Lei, 2010). In contrast, loss of Mael significantly increased red pigment expression for both reporters. Finally, we tested the effect of loss of Mael on the classic PEV allele, white-mottled4 (w^m4), in which a chromosomal inversion moves the euchromatic white gene near centromeric heterochromatin (Muller, 1930). w^m4 flies have white eyes. As expected, both Su(var)2-5 and Su(var)3-9 heterozygotes restored red eye color to w^m4 flies, increasing the amount of red pigment 120- and 150-fold, respectively. Similarly, loss of one copy of mael also suppressed w^m4 silencing, increasing red pigment 80–90 times (Figure S6 and Table S1). Thus, mael is a classical suppressor of PEV. We suggest that loss of Mael suppresses PEV in the somatic cells of the adult eye by the same mechanism it suppresses germline silencing of the gfp reporter in zip and 42AB and of the gypsy12 LTRs in clusters 42AB and 62—by repressing canonical transcription within heterochromatin.

DISCUSSION

Mael Represses Canonical Transcription Activated by Rhi

Fly piRNA clusters must solve a gene-expression paradox. They record the ancient and contemporary exposure of the animal to transposon invasion, and this information must be copied into RNA in order to generate protective, anti-transposon piRNAs. However, recent transposon insertions retain the ability to produce mRNA encoding proteins required for their transposition. In flies, dual-strand piRNA clusters solve this paradox by using Rhi to initiate non-canonical transcription of unspliced RNA from both genomic strands, generating piRNA precursors, while repressing promoter-initiated, canonical transcription (Le Thomas et al., 2014; Mohn et al., 2014; Zhang et al., 2014; Andersen et al., 2017). Our data suggest that Mael is required for this second process, allowing dual-strand piRNA clusters to safely generate piRNA precursor transcripts without risking production of transposon mRNAs (Figure 7). Within dual-strand clusters, Mael is likely guided to its targets by the piRNA pathway. However, our analyses also predict that, for some loci, Mael functions to repress canonical transcription in heterochromatin separately from the piRNA pathway, probably through protein directing Mael to specific genomic sites.

Figure 7. Model for Mael-Dependent Repression of Canonical Transcription.

Dual-strand piRNA cluster transcription is dependent on Rhi and associated proteins, which allow Pol II to initiate independent of promoters and produce non-canonical transcripts. While non-canonical, Rhi-mediated transcription may expose promoters that would otherwise be hidden by heterochromatin. Mael blocks promoter-driven transcription. Without Mael, canonical transcripts are produced from dual-strand clusters and can be translated into protein.

Mael is Required for the Production of Dual-Strand Cluster piRNAs

In mael^M391/r20 ovaries, piRNAs mapping to dual-strand clusters decrease, despite a concomitant increase in canonical transcription from these same loci (Figures 2A and 2B). Our data suggest that in mael^M391/r20 mutants, Rhi-mediated non-canonical transcription, cluster transcript export, and ping-pong amplification become uncoupled. Perhaps, canonical transcription and Rhi-mediated non-canonical transcription compete for Pol II. Instead of fueling piRNA production, the canonical transcripts from dual-strand piRNA clusters produced in the absence of Mael are translated into protein (Figures 2A and 2B). Mael therefore contributes to dual-strand cluster piRNA production by tipping the balance towards non-canonical transcription.

A Putative Conserved Role for Mael

Our data indicate that in addition to relying on the piRNA pathway, Mael can also be guided to its targets by piRNA-independent mechanisms. Moreover, piRNA-dependent repression by Mael may be widespread outside of flies: although Drosophila melanogaster piwi is found only in the gonads, piwi is expressed broadly in the soma of most arthropods (Fu et al., 2018b; Lewis et al., 2018). In fact, 12 arthropods with somatic piRNAs also express mael in the soma, while 3 arthropods with no detectable piRNAs outside the gonads have low or undetectable somatic mael mRNA (Figure S7 and Table S2).

In male mice, loss of MAEL also leads to loss of piRNAs, germline transposon derepression, and sterility (Costa et al., 2006; Soper et al., 2008; Aravin et al., 2009; Castaneda et al., 2014). As in flies, loss of MAEL in mice does not trigger loss of heterochromatin: DNA methylation of L1 elements is unchanged (Aravin et al., 2009).

Because Mael is conserved from protists to humans, we hypothesize that in different organisms Mael may be co-opted by different pathways to repress transcription of various targets. Prior studies suggest a model for how Mael confers repression. In protists, the MAEL domain was predicted to degrade RNA and may directly destroy nascent transcripts (Zhang et al., 2008a). In insects, the MAEL domain interacts with single-stranded RNA (Chen et al., 2015; Matsumoto et al., 2015); we speculate that fly Mael may have retained a role in destabilizing RNA. In this view, Mael may promote premature termination or degradation of nascent transcripts. In addition, because fly Mael has a partial HMG domain (Findley, 2003), it may also directly bind to DNA and repress transcription by preventing canonical core transcription factors from binding to promoters. Another possibility is that fly Mael may have also retained a role in establishing or maintaining chromatin modifications not monitored in this study. Consistent with any of these possible mechanisms, non-canonical transcription mediated by Rhi is expected to be unaffected by Mael because the Rhi allows transcriptional initiation in dual-strand clusters without need for promoters and prevents degradation of unspliced piRNA precursor transcripts.

STAR METHODS

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to, and will be fulfilled by, the Lead Contact, Phillip D. Zamore (Phillip.Zamore@umassmed.edu).

METHOD DETAILS

Drosophila Stocks

Fly stocks were maintained at 25°C. All strains were in the w¹¹¹⁸ background except w⁺; rhi^KG. Both the mael^M391 and mael^r20 alleles were backcrossed to w¹¹¹⁸ for five generations before use to minimize genetic background effects. The armi^G728E allele was outcrossed for six generations, and the FRT site sequences found in the original line removed. The P{GSV6}42A18 and P{GSV6}42A18 transgenes derive from P{GSV6}^Gs13456 (Toba et al., 1999) and are located at Chr2R: 6,460,398-6,460,415 and Chr2R: 25,012,839-25,013,065, respectively (dm6).

General Methods

Before dissection, flies were isolated 0–3 days after eclosion and given yeast paste for two days. Fly ovaries were then dissected and collected in 1× phosphate-buffered saline [pH 7.4] (1×PBS) (137 mM NaCl, 2.7 mM KCl, 10 mM Na₂HPO₄, 1.8 mM KH₂PO₄) cooled on ice. Ovaries were then washed once with ice-cold 1×PBS and then used for subsequent experiments.

Western Blotting

Ovary lysate was prepared as described (Li et al., 2009a) with modifications. After 1×PBS was removed, the ovaries were homogenized with a plastic pestle (Fisher Scientific, #12141364) in ice-cold lysis buffer (for each 100 mg ovaries, 100 μl of 100 mM potassium acetate, 30 mM HEPES-KOH [pH 7.4], 2 mM magnesium acetate, 1 mM dithiothreitol (DTT)) containing 1 mM AEBSF (4-(2-aminoethyl)benzenesulfonyl fluoride hydrochloride; EMD Millipore, #101500), 0.3 μM Aprotinin (Bio Basic Inc, #AD0153), 20 μM Bestatin (Sigma Aldrich,#B8385), 10 μM E-64 ((1S,2S)-2-(((S)-1-((4-Guanidinobutyl)amino)-4-methyl-1-oxopentan-2-yl)carbamoyl)cyclopropanecarboxylic acid; VWR, #97063), and 10 μM Leupeptin (Fisher Scientific, #108975). Lysate was centrifuged at 13,000 × g for 30 min at 4°C and an equal volume of 2× loadin g dye (100 mM Tris-HCl [pH 6.8], 4% (w/v) SDS, 0.2% (w/v) bromophenol blue, 20% (v/v) glycerol, and 200 mM DTT) was added to the supernatant and heated to 95°C for 5 min. The lysate was resolved through a 4–20% gradient polyacrylamide/SDS gel electrophoresis (Bio-Rad Laboratories, #5671085). After electrophoresis, proteins were transferred to a 0.45 μm pore polyvinylidene difluoride membrane (Millipore, #IPVH00010), the membrane blocked in Blocking Buffer (Rockland Immunochemicals, #MB-070) at room temperature for 1 h and then incubated overnight at 4°C in 1:1 Blocking Buffer: 1× TBST [pH 7.5] (50 mM Tris-HCl [pH 7.5], 150 mM NaCl, 0.1% Tween 20 (v/v)) containing primary antibody (anti-GFP, Santa Cruz Biotechnology, #SC-9996, 1:2500 dilution; anti-α-Tubulin, DSHB, #12G10, 1:50,000 dilution, anti-Mael, gift from Julius Brennecke, 1:2500 dilution). The membrane was washed three time, each for 5 min, with 1× TBST [pH 7.5] at 25°C, incubated in Blocking Buffer dilu ted 1:1 in 1× TBST [pH 7.5] and containing secondary antibody (donkey anti-mouse IRDye 680RD, LICOR Biosciences, #926-68072, 1:10,000 dilution; goat anti-mouse IRDye 800CW, LICOR Biosciences, #926-32210, 1:10,000 dilution) for 1 h at room temperature in the dark, and washed five times for 10 min each with 1× TBST [pH 7.5] at room temperature in the dark. Signal was detected using an Odyssey Infrared Imaging System. Data were obtained from three independent biological replicates. Quantification was performed using Image Studio v4.0.21 (LI-COR). p-values were measured using an unpaired, two-tailed t-test.

Immunohistochemistry and Microscopy

Immunohistochemistry and microscopy was performed as described (Li et al., 2009a). After ovaries were teased apart using a pipette, they were fixed in 4% formaldehyde in 15 mM piperazine-N,N′-bis(2-ethanesulfonic acid) (PIPES) [pH 7.4], 80 mM KCl, 20 mM NaCl, 2 mM EDTA, 0.5 mM EGTA, and 1 mM DTT (Buffer A) for 10 min, rotating at room temperature. Afterwards, ovaries were washed twice for 15 min each at room temperature in Buffer A with 0.1% (w/v) Triton X-100. The ovaries were washed again twice for 15 min each at room temperature in Buffer A with 10% (v/v) normal donkey serum (Sigma, #D9663) and 0.1% (w/v) Triton X-100. The ovaries were then incubated with anti-Rhi (gift from William Theurkauf Zhang et al., 2014; 1:1000 dilution) in the previous buffer rotating at 4°C overnight.

The next day, the ovaries were washed four times for 30 min each at room temperature in Buffer A with 2 mg/ml BSA and 0.1% (w/v) Triton X-100, followed by a 30 min wash at room temperature in Buffer A with 10% (v/v) normal donkey serum and 0.1% (w/v) Triton X-100. The ovaries were then incubated in the dark, rotating overnight at 4°C with secondary antibody (Donkey anti-Guinea Pig IgG, CF594, Sigma-Aldrich, #SAB460009, 1:100 dilution) in Buffer A with 10% (v/v) normal donkey serum and 0.1% (w/v) Triton X-100.

The next day, ovaries were washed twice for 30 min each at room temperature in in Buffer A with 0.1% (w/v) Triton X-100, followed by a 1 min wash at room temperature with Buffer A. Ovaries were then mounted using VECTASHIELD Mounting Medium (Vector Laboratories, #H-1000) and covered with a 0.13–0.17 mm thick cover slip (VWR #48393 106). Images were captured using a Leica TCS SP5 II Laster Scanning Confocal Microscope using the 63× objective with Olympus Immersion Oil Type-F (Thorlabs #MOIL-30). Images presented in the same figure were acquired at the same settings.

Female Fertility Assay

Female fertility was tested essentially as described (Li et al., 2009a). Eight female virgins were mated to four Oregon R virgin males in a small cage with a 60 mm diameter grape juice agar plate dabbed with yeast paste at 25°C. All flies were collected <1-day post-eclosion. After two days, the first plate was discarded and replaced with a fresh plate. Plates were then changed and scored every subsequent day. The number of total eggs, eggs per female per day, and the dorsal appendage phenotype of embryos were scored every 24 h and the number of eggs that hatched were scored 48 h after the plate was changed. Fertility was recorded for 12 days and at least two independent biological replicates were conducted for each genotype.

Eye Pigment Assay

Ethanol-based pigment extraction and quantification was performed essentially as described (Sun et al., 2004). Briefly, 10 females, 3–5 days post-eclosion, were collected and their heads photographed and dissected. Heads were homogenized in 0.5 ml of 0.01 M HCl in ethanol. The homogenate was incubated at 4°C rotating overnight, warmed to 50°C for 5 min, and centrifuged at 13,000 × g for 10 min at 25°C. The supernatant was collected, and absorbance at 480 nm (A480) was measured. Background from w¹¹¹⁸ female heads was subtracted from each measurement. p-values were measured using an unpaired, two-tailed t-test.

Construction and Analysis of sRNA-Seq Libraries

Small RNA libraries were constructed as described (Han et al., 2015). In summary, total RNA (50 μg) was extracted using mirVana miRNA Isolation Kit (Life Technologies, #AM1561) and purified by 15% urea polyacrylamide gel electrophoresis (PAGE), selecting for 18–30 nt small RNAs. Half of the purified sRNAs were oxidized with NaIO₄ was used to deplete miRNAs and enrich for siRNAs and piRNAs (Li et al., 2009a). To reduce ligation bias, a 3′ adaptor with three random nucleotides at its 5′ end was used (5′-rApp NNN TGG AAT TCT CGG GTG CCA AGG /ddC/-3′). 3′ adaptor was ligated using truncated, K227Q mutant T4 RNA Ligase 2 (made in lab) at 25°C for ≥16 h, sRNAs precipitated, and size selected as described in (Li et al., 2009a). To exclude 2S rRNA from sequencing libraries, 10 pmol 2S blocker oligo was added before 5′ adaptor ligation (Wickersheim and Blumenstiel, 2013). 5′ adaptor was added using T4 RNA ligase (Life Technologies, #AM2141) at 25°C for 2.5 h, followed by reverse-transcription using AMV reverse transcriptase (New England Biolabs, #M0277L) and PCR using AccuPrime Pfx DNA Polymerase (Invitrogen, #12344-024). Small RNA-seq libraries for three independent biological replicates were sequenced using a NextSeq500 (Illumina) to obtain 75 nt single-end reads.

sRNA-seq analysis was performed with piPipes using default options (v1.4; Han et al., 2014). Briefly, barcodes were sorted allowing one mismatch, and the 3′ adaptors, including the three random nucleotides, were identified and removed using the first ten nucleotides, allowing one mismatch. After adaptor removal, reads containing one or more nucleotides with Phred score <5 were discarded. sRNAs were first aligned to rRNA or miRNA hairpin sequences using Bowtie allowing 2 and 0 mismatches respectively(v1.2.0; Langmead et al., 2009). Unaligned reads were mapped to the genome (using the options ‘-r -v 0 -a --best --strata’) and 23–29 nt RNAs passing SAMtools filtering (v1.9; ‘samtools view -uS -F0×4’; Li et al., 2009b) were kept for analyses. The number of piRNAs overlapping each genomic feature (genes, transposons, and piRNA producing loci) were apportioned by the number of times they aligned to the genome.

Oxidized sRNA libraries are enriched with piRNAs. Therefore, to compare piRNA abundances across different oxidized libraries, we calibrated oxidized to unoxidized libraries. Because paired oxidized and unoxidized sRNA libraries were created from the same source, the subset of piRNA species should remain constant between the two libraries. First, unoxidized libraries were normalized to sequencing depth (ppm). Next, we identified all the uniquely mapping piRNA species (piRNAs that shared the exact nucleotide sequences) that were shared between at least two of the three replicates of paired oxidized and unoxidized libraries. Finally, the calibration factor was computed using the ratio between the sums of the normalized abundance in the unoxidized libraries and the abundances in the oxidized libraries,

$$\mathit{calibration}\mathit{factor}=\frac{{\sum }_{c}pp{m}_{c,unox}}{{\sum }_{c}count{s}_{c,ox}}$$

where c is the number of common piRNA species between oxidized and unoxidized libraries. piRNAs in the top 10^th percentile were excluded to avoid overweighting outliers. The abundance of each piRNA in the oxidized library was calculated by multiplying by the calibration factor. Smoothing of the data points with non-parametric regression (LOWESS) was conducted in R without robustifying iterations and the span set at 10 nt.

Ping-Pong Analysis

Ping-pong analysis was as described (Zhang et al., 2011). In summary, scores at each 5’-to-5’ distance for two piRNAs were defined as the product of their abundances. The Ping-Pong Z₁₀ score was then the difference of the score at the 5’-to-5’ distance of 10 nt and the mean scores of background distances, divided by the standard deviation of the scores of background distances, defined as distances of 0–9 and 11–20 nt. For analyses including multiply mapping reads, read abundances were apportioned by the number of times the read aligned to the genome.

Phasing Analysis

Phasing analysis was as described (Han et al., 2015). Briefly, sRNA reads were mapped to genome and rRNAs, tRNAs, and snoRNAs were removed. The Z_x score for a distance x between the 3′ end of one piRNA to the 5′ end of a downstream piRNA on the same genomic strand was calculated by the difference of the score at the distance x and the mean scores of background distances, divided by the standard deviation of the scores of background distances. When x = 0, the 5′ end is immediately downstream of the 3′ end (phasing). For analyses including multiply mapping reads, read abundances were apportioned by the number of times the read aligned to the genome. Overlaps at positions 1–20 were used as background to calculate Z₀.

Construction and Analysis of RNA-Seq Libraries

RNA-seq libraries were constructed as described (Zhang et al., 2012) with several modifications, including the use of unique molecular identifiers to eliminate PCR duplicates (Fu et al., 2018a). For ribosomal RNA depletion, RNA was hybridized in 10 μl with a pool of 186 rRNA antisense oligos (0.05 μM/each; Morlan et al., 2012; Adiconis et al., 2013) in 10 mM Tris-HCl [pH 7.4] and 20 mM NaCl and heated to 95°C, then cooled at −0.1°C/sec to 22°C, and finally incubated at 22°C f or 5 min. Ten units of RNase H (Lucigen, #H39500) were added and incubated at 45°C for 30 min in 20 μl containing 50 mM Tris-HCl [pH 7.4], 100 mM NaCl, and 20 mM MgCl₂. RNA was then treated with 4 units DNase (Thermo Fisher, #AM2238) in 50 μl at 37°C for 20 min. After DNase treatment, RNA was purified using RNA Clean & Concentrator-5 (Zymo Research, #R1016). RNA-seq libraries were sequenced using a NextSeq500 (Illumina) to obtain 75 + 75 nt, paired-end reads.

RNA-seq analysis was performed with piPipes using default options(v1.4; Han et al., 2014). Briefly, barcodes were sorted allowing one mismatch and were identified and removed using the first ten nucleotides, allowing one mismatch. RNAs were first aligned to rRNA sequences using Bowtie2 (v2.2.0; Langmead and Salzberg, 2012) using the following parameters ‘--very-fast –no-mixed –no-discordant -k1 ‘. Unaligned reads were then mapped using STAR to the fly genome (v2.3.1; Dobin et al., 2013) with the following options ‘--runMode alignReads --limitOutSAMoneReadBytes 1000000 -- outFilterScoreMin 0 --outFilterScoreMinOverLread 0.72 --outFilterMatchNmin 0 -- outFilterMatchNminOverLread 0.72 --outFilterMultimapScoreRange 1 -- outFilterMultimapNmax -1 --outFilterMismatchNmax 10 --outFilterMismatchNoverLmax 0.05 --alignIntronMax 0 --alignIntronMin 21 --outFilterIntronMotifs RemoveNoncanonicalUnannotated --genomeLoad NoSharedMemory -- outSAMunmapped None --outReadsUnmapped Fastx --outSJfilterReads Unique -- seedSearchStartLmax 20 --seedSearchStartLmaxOverLread 1.0 --chimSegmentMin 0’. Counts were produced using the “strict” option on HTseq (v0.6.1; Anders et al., 2015). The genomic coordinates of the left and right promoter regions flanking cluster 38C1, and the region within the cluster are Chr2L: 20,104,764-20,105,180, Chr2L: 20,115,547-20,115,964, and Chr2L: 20,105,180-20,115,547, respectively (dm6).

Construction and Analysis of ChIP-Seq Libraries

ChIP-seq libraries were constructed as described (Zhang et al., 2014) with several modifications. Briefly, ~100 μl ovaries per library were first crosslinked with 2% formaldehyde for 10 min rotating at 25°C in Robb’s medium (100 mM HEPES [pH 7.4], 55 mM sodium acetate, 40 mM potassium acetate, 100 mM sucrose, 10 mM glucose, 1.2 mM MgCl₂, 1 mM CaCl₂, 1mM DTT, 1 mM AEBSF, 0.3 μM Aprotinin, 20 μM Bestatin, 10 μM E-64, and 10 μM Leupeptin). Crosslinking was quenched by adding Glycine to a final concentration of 120 mM and for 5 min rotating at 25°C. The ovaries were then washed twice with TBS (50 mM Tris-HCl [pH 7.5], 150 mM NaCl), and twice with ChIP lysis buffer (50 mM HEPES-KOH [pH 7.5], 140 mM NaCl, 1% [v/v] Triton X-100, 0.1% [w/v] Na-Deoxycholate, 0.1% [w/v] SDS).

Ovaries were then sonicated in sonication buffer (1% [w/v] SDS, 10 mM EDTA, 50 mM Tris-HCl [pH 8.0], 1mM DTT, 1 mM AEBSF, 0.3 μM Aprotinin, 20 μM Bestatin, 10 μM E-64, and 10 μM Leupeptin) using an E220 Evolution Focused-ultrasonicator (Covaris) with Duty cycle: 5%, Intensity: 140 watts, Cycles per burst: 200, Temperature: <10°C, Time: 20 min. The sonicated lysate was centrifuged at 13,000 χ g for 15 min at 4°C. Supernatant was diluted 7-fold with dilution buffer (20 mM Tris-HCl [pH 7.5], 167 mM NaCl, 1.2 mM EDTA, 0.01% [w/v] SDS, 1% [v/v] Triton X-100, 1mM DTT, 1 mM AEBSF, 0.3 μM Aprotinin, 20 μM Bestatin, 10 μM E-64, and 10 μM Leupeptin) and incubated overnight rotating at 4°C with antibody (anti-Rhi or Pre-Immune Serum, gift from William Theurkauf, 20 μl; anti-HP1a, DSHB, #C1A9, 5 μg; normal mouse IgG, Abcam, #ab188776, 5 μg; anti-H3K9me3, Abcam, #ab8898, 10.5 μg; anti-H3K4me3, Abcam, #ab8580, 10.5 μg; anti-Histone H3, Abcam #ab18521, 10.55 μg) bound to 100 μl of Dynabeads Protein A/G (Life Technologies, #10002D/#10004D).

The beads were then washed 2×5 min each with 500 μl of the following buffers: Wash buffer A (20 mM Tris-HCl [pH 8.0], 2 mM EDTA, 0.1% [w/v] SDS, 1%[v/v] Triton X-100, 150 mM NaCl), Wash buffer B (20 mM Tris-HCl [pH 8.0], 2 mM EDTA, 0.1% [w/v] SDS, 1%[v/v] Triton X-100, 500 mM NaCl), Wash buffer C (10 mM Tris-HCl [pH 8.0], 1 mM EDTA, 1% [v/v] NP-40, 1% [w/v] Na-deoxycholate, 0.25 M LiCl) and Wash buffer D (10 mM Tris-HCl [pH 8.0], 1 mM EDTA). All wash buffers also contained 1mM DTT, 1 mM AEBSF, 0.3 μM Aprotinin, 20 μM Bestatin, 10 μM E-64, and 10 μM Leupeptin. Beads were then treated with 20 μg/ml RNase A (Fisher Scientific, #FEREN0531) To reverse crosslink and remove protein, beads were incubated overnight at 65°C with 200 μg/ml Proteinase K (Life Technologies, #25530015) in 2×Proteinase K Buffer (200 mM Tris-HCl [pH 7.5], 2 mM EDTA [pH 8.0], and 1% SDS (w/v). Finally, DNA was purified using phenol:chloroform [pH 8.0] and the library was prepared by sequentially performing end-repair, A-tailing, Y-shaped adaptor ligation, and PCR amplification as described (Zhang et al., 2012).

Barcodes were sorted allowing one mismatch and were identified and removed using the first ten nucleotides, allowing one mismatch. For H3K9me3, HP1a, and Rhi ChIP-seq libraries, reads were processed using piPipes using default options(v1.4;). Briefly, reads were mapped on the genome using Bowtie2 (v2.2.0; Langmead and Salzberg, 2012) with the following options (‘--very-sensitive-local -X 800 --no-mixed’). Unmapped reads were removed using SAMtools (v0.1.19; Li et al., 2009b; Li, 2011) and a mapping q-score of 10 was used to identify uniquely mapping reads. We used a 1 kbp sliding window with a 500 bp step over the genome to compute a signal for each chromosome arm. For each chromosome arm, the counts for each bin were normalized using the total number of reads mapping to the chromosome. Counts were produced using the “strict” option on HTseq (v0.6.1; Anders et al., 2015). Reads were normalized to sequencing depth.

For H3K4me3 ChIP-seq libraries, reads were mapped to the genome using BWA (Li and Durbin, 2009). SAMtools (v0.1.19; Li et al., 2009b; Li, 2011) and Picard Toolkit (Toolkit)Toolkit were used to remove improperly paired reads and PCR duplicates. MACS2 (Zhang et al., 2008b) was used to called peaks. We used BEDtools (v2.26.0; Quinlan and Hall, 2010) to merge peaks from all replicates for each genotype to create a consensus set of peaks. The number of reads overlapping each peak was computed using BEDtools and reads were normalized to sequencing depth. The H3K4me3 signal around TSS of genes was performed using the deepTools software (v3.1.2; Ramírez et al., 2016). Firstly, we used the function computeMatrix, using the following options ‘reference-point -b 5000 -a 5000 --referencePoint TSS -skipZeros’, to produce the signal around the TSS. The plot were produced using the function plotHeatmap with the following parameters ‘--colorMap Blues --missingDataColor 1 --sortRegions descend -- sortUsingSamples 3 4’. We repeated the same approach to compute the signal around TE insertions using their 5 primes as reference point.

Construction and Analysis of GRO-Seq Libraries

GRO-seq libraries were constructed as described (Wang et al., 2015), and analyzed with piPipes like describe above for RNA-seq (v1.4; Han et al., 2014). Briefly, 0–2-day-old female flies were given yeast for 2 days before their ovaries were dissected. One hundred pairs of ovaries were homogenized in 350 μl HB35 buffer (15 mM HEPES KOH [pH 7.5], 10 mM KCl, 2.5 mM MgCl₂, 0.1 mM EDTA, 0.5 mM EGTA, 0.05% [v/v] NP 40, 0.35 M sucrose, 1 mM DTT, 1 mM AEBSF, 0.3 μM Aprotinin, 20 μM Bestatin, 10 μM E-64, and 10 μM Leupeptin) with a Dounce homogenizer using pestle B (Sigma Aldrich, #D8938). Nuclei were purified by passing twice through sucrose cushions that contain 800 μL HB80 buffer (15 mM HEPES KOH [pH 7.5], 10 mM KCl, 2.5 mM MgCl₂, 0.1 mM EDTA, 0.5 mM EGTA, 0.05% [v/v] NP 40, 0.80 M sucrose, 1 mM DTT, 1 mM AEBSF, 0.3 μM Aprotinin, 20 μM Bestatin, 10 μM E-64, and 10 μM Leupeptin) on the bottom phase and 350 μl HB35 buffer on the top. Nuclei were washed once with 500 μl freezing buffer (50 mM Tris-HCl, pH 8.0, 40% [v/v] glycerol, 5 mM MgCl2, 0.1 mM EDTA, 1 mM dithiothreitol, 1 mM AEBSF, 0.3 μM Aprotinin, 20 μM Bestatin, 10 μM E-64, and 10 μM Leupeptin) and frozen in liquid nitrogen with 100 μL freezing buffer. To carry out nuclear run on assay, 100 μL freshly prepared reaction buffer (10 mM Tris-HCl, pH 8.0, 5 mM MgCl2, 300 mM KCl, 1% [w/v] sarkosyl, 500 μM ATP, 500 μM GTP, 500 μM Br UTP, 2.3 μM CTP, 1 mM dithiothreitol, 20 U RNasin Plus RNase Inhibitor (Promega, #N2615) was added to nuclei and incubated at 30°C for 5 min . RNA was extracted using Trizol (Invitrogen, #15596). Nascent RNAs with Br UTP incorporated were enriched by immunoprecipitation using anti 5 bromo 2′ deoxyuridine antibody (Fisher Scientific, #50175223) as described (Shpiz and Kalmykova, 2014), followed by rRNA depletion using RNase H, fragmentation, and library construction as in RNA-seq library preparation (Zhang et al., 2012).

Analysis of Mael Orthologs in Arthropods

Mael orthologs in arthropods were identified using either available genome annotation (Table S2; Musca domestica, Plutella xylostella, Nicrophorus vespilloides, Limus polyphemus, Parasteatoda tepidariorum, Centruroides sculpturatus, Trichoplusia ni) or OrthoDB v9.1 (group EOG091G0BYM; Zdobnov et al., 2017; Drosophila virilis, Aedes aegypti, Tribolium castaneum, Apis mellifera, Bombus terrestris, Oncopeltus fasciatus, Acyrthosiphon pisum). RNA-seq data sets (Table S2; Lewis et al., 2018; Fu et al., 2018b) were first aligned to the corresponding organism’s ribosomal RNA sequences (SILVA database; Quast et al., 2013) using Bowtie2 (v2.2.0; Langmead and Salzberg, 2012). Unaligned reads were then mapped to the corresponding organism’s genome assembly (Table S2) using STAR (v2.3.1; Dobin et al., 2013). Sequencing depth and gene quantification was calculated with Cufflinks (v2.1.1; Trapnell et al., 2010).

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse monoclonal anti-GFP	Santa Cruz Biotechnology	Cat# SC-9996 RRID:AB_627695
Mouse monoclonal anti-alpha-tubulin	DSHB	Cat# 12G10 RRID:AB_1157911
Mouse anti-Maelstrom	J. Brennecke	N/A
Guinea-pig anti-Rhino	(Zhang et al., 2014)	N/A
IRDye 680RD Donkey anti-mouse	LI-COR Biosciences	Cat# 926-68072 RRID:AB_10953628
IRDye 800CW Goat anti-mouse	LI-COR Biosciences	Cat# 926-32210 RRID:AB_621842
Donkey anti-Guinea Pig IgG, CF594	Sigma-Aldrich	Cat# SAB4600096 RRID: AB_2728625
Deposited Data
Raw sequencing data from D. melanogaster	This paper	SRA: PRJNA448445
Experimental Models: Organisms/Strains
D. melanogaster: mael mutant: w1118; +; maelM391	(Clegg et al., 1997)	N/A
D. melanogaster: mael mutant: w1118; +; maelr20	Bloomington Drosophila Stock Center	BDSC:8516 RRID:BDSC_8516
D. melanogaster: rhi mutant: w1118; P{PZ}rhi02086 (rhi2); +	Bloomington Drosophila Stock Center	BDSC:12226 RRID:BDSC_12226
D. melanogaster: rhi mutant: w+; P{SuPor-P}rhiKG00910 (rhiKG); +	Bloomington Drosophila Stock Center	BDSC:13161 RRID:BDSC_13161
D. melanogaster: RNAi of Rhi: w1118; +; P{TRiP.GL00041}attP2	Bloomington Drosophila Stock Center	BDSC:35171 RRID:BDSC_35171
D. melanogaster: RNAi of Cuff: w1118; +; P{TRiP.GL00054}attP2	Bloomington Drosophila Stock Center	BDSC:35182 RRID:BDSC_35182
D. melanogaster: piwi mutant: w1118; P{ry11}piwi2; +	Bloomington Drosophila Stock Center	BDSC:43319 RRID:BDSC_43319
D. melanogaster: piwi mutant: w1118; P{lacZ}piwiNt; +	(Klenov et al., 2011)	N/A
D. melanogaster: armi mutant: w1118; +; armi72.1	Bloomington Drosophila Stock Center	BDSC:8544 RRID:BDSC_8544
D. melanogaster: armi mutant: w1118; +; armiG728E	(Hayashi et al., 2014)	N/A
D. melanogaster: P{GSV6}42A18: w1118; P{GSV6}42A18; +	(Han et al., 2015)	N/A
D. melanogaster: P{GSV6}zip: w1118; P{GSV6}zip; +	(Han et al., 2015)	N/A
D. melanogaster: wm4 mutant: ln(1)w[m4]; +; +	Bloomington Drosophila Stock Center	BDSC:807 RRID:BDSC_807
D. melanogaster: mini-white insertion: y[1] w[67c23] P{w[+mC] y[+mDint2]=EPgy2}DIP1[EY02625]	Bloomington Drosophila Stock Center	BDSC:15577 RRID:BDSC_15577
D. melanogaster: mini-white insertion: w1118; P{w[+mC] y[+mDint2]=EPgy2}EY08366; +	Bloomington Drosophila Stock Center	BDSC:19874 RRID:BDSC_19874
D. melanogaster: UASp-FLAG-Mael rescue: w1118; +; UASp-FLAG-Mael, maelM391	(Pek et al., 2009)	N/A
D. melanogaster: nanos-GAL4 driver: w1118; P{GAL4-nanos.NGT}40; +	Bloomington Drosophila Stock Center	BDSC:4442 RRID:BDSC_4442
D. melanogaster: MTD-GAL4 driver: P{w[+mC]=otu-GAL4::VP16.R}1, w[*]; P{w[+mC]=GAL4-nanos.NGT}40; P{w[+mC]=GA L4:: VP16-n an os. UTR}CG6325[MVD1]	Bloomington Drosophila Stock Center	BDSC:31777 RRID:BDSC_31777
Chemicals, Peptides, and Recombinant Proteins
VECTASHIELD Mounting Medium	Vector Laboratories	Cat# H-1000 RRID:AB_2336789
Software and Algorithms
Bowtie2	(Langmead and Salzberg, 2012)	http://bowtie-bio.sourceforge.net/bowtie2/index.shtml
STAR	(Dobin et al., 2013)	https://code.google.com/archive/p/rna-star/
piPipes	(Han et al., 2014)	http://bowhan.github.io/piPipes/
HTseq	(Anders et al., 2015)	https://pypi.org/project/HTSeq/
SAMtools	(Li et al., 2009b)	http://samtools.sourceforge.net
BWA	(Li and Durbin, 2009)	http://maq.sourceforge.net
BEDtools	(Quinlan and Hall, 2010)	https://code.google.com/archive/p/bedtools/
Picard Toolkit	N/A	http://broadinstitute.github.io/picard/
MACS2	(Zhang et al., 2008b)	https://github.com/taoliu/MACS/wiki

Highlights.

• Maelstrom represses canonical transcription within fly dual-strand piRNA clusters

• Maelstrom also represses canonical transcription in conventional heterochromatin

• Maelstrom can be guided to its targets both by piRNAs and piRNA-independent mechanisms