PIWI Slicing and EXD1 Drive Biogenesis of Nuclear piRNAs from Cytosolic Targets of the Mouse piRNA Pathway.

PIWI-interacting RNAs (piRNAs) guide PIWI proteins to suppress transposons in the cytoplasm and nucleus of animal germ cells, but how silencing in the two compartments is coordinated is not known. Here we demonstrate that endonucleolytic slicing of a transcript by the cytosolic mouse PIWI protein MILI acts as a trigger to initiate its further 5'→3' processing into non-overlapping fragments. These fragments accumulate as new piRNAs within both cytosolic MILI and the nuclear MIWI2. We also identify Exonuclease domain-containing 1 (EXD1) as a partner of the MIWI2 piRNA biogenesis factor TDRD12. EXD1 homodimers are inactive as a nuclease but function as an RNA adaptor within a PET (PIWI-EXD1-Tdrd12) complex. Loss of Exd1 reduces sequences generated by MILI slicing, impacts biogenesis of MIWI2 piRNAs, and de-represses LINE1 retrotransposons. Thus, piRNA biogenesis triggered by PIWI slicing, and promoted by EXD1, ensures that the same guides instruct PIWI proteins in the nucleus and cytoplasm.

Introduction

Transposons are mobile genetic elements that have the ability to move within the genome. Uncontrolled transposition is linked to genome damage, so it is imperative that all active elements are silenced. Animal gonads express a set of 24–30 nt small RNAs called PIWI-interacting RNAs (piRNAs) that together with their partner PIWI protein constitute a transposon defense system (Houwing et al., 2007). Sequence complementarity to host transposable elements allows piRNAs to guide associated PIWI proteins to target nucleic acids. Cytoplasmic PIWI proteins are small RNA-guided nucleases (slicers) that guide endonucleolytic cleavage of transposon targets, while nuclear PIWI proteins assemble silencing complexes on target genomic loci to mediate transcriptional silencing (Ghildiyal and Zamore, 2009).

The majority of piRNAs originate from RNA polymerase II transcription units called piRNA clusters. These are transcribed into 50- to 100-kb-long capped and polyadenylated precursor transcripts (Li et al., 2013). Precursors are then exported to the cytoplasm for processing as most known piRNA biogenesis factors are cytoplasmic proteins and are resident in perinuclear granules called nuage. The single-stranded cluster transcripts are then converted into tens of thousands of piRNAs via a process termed primary processing (Brennecke et al., 2007). The property that recruits the piRNA precursors into the primary processing pathway lies within the transcripts themselves. Drosophila piRNA precursors were shown to contain non-sequence conserved cis-acting RNA elements (Ishizu et al., 2015) or piRNA trigger sequences (PTSs) (Homolka et al., 2015) that, when fused to any heterologous transcript, can direct them into primary processing. These elements provide binding sites for piRNA biogenesis factors (Ishizu et al., 2015) and act in a directional (5′→3′) manner by converting any sequence downstream into thousands of primary piRNAs with a prominent bias for carrying a uridine at the 5′ end (1U). Thus, the PTS-initiated primary piRNA biogenesis allows the de novo creation of the original pool of piRNAs as instructed by the genetic design of the precursor transcripts. Although primary biogenesis pathway and factors involved are conserved, it remains to be seen whether the use of such signals to initiate processing is a common feature across the animal kingdom.

PIWI slicing is a second mechanism by which piRNA biogenesis can be initiated on a target transcript. Initially described in the fly ovarian germline, cytosolic PIWI endonucleolytic action generates two target cleavage fragments, of which the one downstream from the cleavage site enters the piRNA biogenesis pathway to mature as a new (secondary) piRNA whose 5′ end is created by PIWI slicing (Brennecke et al., 2007, Gunawardane et al., 2007). Recent studies indicate that the downstream cleavage fragment is a source for even more extensive piRNA generation (Han et al., 2015, Homolka et al., 2015, Mohn et al., 2015). It was found that an additional series of non-overlapping piRNAs are generated from the fragment in a 5′→3′ direction, a process we termed inchworming (Homolka et al., 2015). The generated piRNAs display a strong preference for 1U, a primary piRNA feature, demonstrating that PIWI slicing serves to direct a target of the piRNA pathway into primary processing. Interestingly, the primary piRNAs generated by inchworming are exclusively loaded into a fly nuclear PIWI protein, thereby linking post-transcriptional silencing in the cytoplasm to increasing information on that particular target in the nuclear compartment. A similar functional relationship exists between two mouse PIWI proteins. Transposon silencing in the mouse embryonic male germline is secured by the slicer-competent cytoplasmic MILI (Di Giacomo et al., 2013) and the slicer-inactive nuclear MIWI2 that is proposed to recruit transcriptional silencing complexes on genomic transposon loci (De Fazio et al., 2011). It is known that MILI (Aravin et al., 2008, Kuramochi-Miyagawa et al., 2008) and its slicer activity (De Fazio et al., 2011) is critical for biogenesis of most of the piRNAs associating with MIWI2, but the molecular mechanisms involved in this process are currently not known.

The repetitive nature of piRNA populations present in the mouse embryonic testes prevents an unambiguous assignment of endogenous piRNAs associating with MILI or MIWI2. To overcome this, we created a mouse knockin line expressing a reporter transcript that is a target for MILI slicing and tracked piRNA biogenesis initiated by MILI slicer activity in vivo. We demonstrate that MILI slicing triggers the generation of non-overlapping 1U-containing piRNAs in a 5′→3′ direction, and these are loaded into both MILI and MIWI2. In addition, we describe the identification and characterization of Exonuclease domain-containing 1 (EXD1) as a partner of the known MIWI2 piRNA biogenesis factor TDRD12. Crystal structures demonstrate that Exd1 exists as homodimers and is inactive as a nuclease. However, the protein functions in an RNA-binding role within a PIWI-Exd1-Tdrd12 (PET) complex. Significantly, loss of Exd1 results in reduced loading of MIWI2 and its consequent retention in the cytoplasm. Sequence analysis indicates that piRNAs generated by MILI slicing are reduced in the mutant. This leads to a reduction in abundance of antisense repeat piRNAs targeting active retrotransposons like LINE1, which become de-repressed in the Exd1 mutant mice. We propose that while PIWI slicing expands the repertoire of nuclear piRNAs by initiating the processing of cytosolic targets of the piRNA pathway, EXD1 ensures the efficient entry of small RNAs into the nuclear PIWI protein.

Results

MILI Slicing Triggers Biogenesis of Non-Overlapping MIWI2 piRNAs in a 5′→3′ Direction

We created a mouse knockin line (hereafter referred to as Rosa26-pi) that expresses a reporter targeted by multiple MILI-bound piRNAs (Figures S1A–S1C; Supplemental Experimental Procedures). We then prepared small RNA libraries of MILI and MIWI2 complexes isolated from testes of new-born pups (post-natal day 0; P0) of the Rosa26-pi genotype (Table S1). Analysis indicates presence of reporter-derived reads in both MILI and MIWI2 libraries, with many originating from within the region predicted to be cleaved by MILI (gray shaded regions in Figure 1A). PIWI proteins slice a target RNA at a fixed distance (10 nt) downstream from the 5′ end of the piRNA guide (De Fazio et al., 2011, Reuter et al., 2011). Indeed, majority of the reporter-derived piRNAs have their 5′ ends at a distance (position −10) consistent with generation by MILI slicing (Figures 1B and S1D). Additionally, since the targeting MILI piRNAs that we chose in the reporter design have a uridine as the first residue (1U), all the reporter-derived reads have an adenosine (A10) at the tenth position (Figure S1E). By definition (Brennecke et al., 2007), these MILI- and MIWI2-loaded sequences are secondary piRNAs as a PIWI protein generates their 5′ ends.

Figure 1

MILI Slicing of a Target Triggers Production of MIWI2 piRNAs in a 5′→3′ Direction

(A) The ubiquitous Rosa26 promoter-driven DsRed reporter transcript has a 3′ UTR with 35 perfectly complementary binding sites (shaded regions) for different MILI-bound piRNAs. See also Figure S1. Slicer cleavage of the transcript by MILI leads to accumulation of reporter-derived piRNAs in both MILI and MIWI2. The 5′ ends of such piRNAs were mapped along the transcript, and those exceeding the abundance of 1 rpm were plotted. The abundance is shown in logarithmic scale.

(B) Mapping of the 5′ and 3′ ends of reporter-derived artificial piRNAs relative to the 5′ end (nucleotide position −1) of the targeting MILI piRNA. Read counts were aggregated for the 35 individual MILI-targeted sites. The y axis was scaled to allow inspection of even low abundant reads. The 5′ ends of secondary piRNAs (at position −10) are the most abundant and are generated by MILI slicing. Another distinct set of 5′ ends (inchworm piRNAs) map downstream from the 3′ end of these secondary piRNAs (∼at positions +18 to +33).

(C) The artificial piRNAs are divided into categories based on the distance of their 5′ ends from that of the targeting piRNA. The graph shows the proportion of individual categories in MILI and MIWI2 libraries.

(D) The pie charts show the 5′ nucleotide preference for different categories of produced piRNAs and the nucleotide composition of the reporter sequence giving rise to them.

(E) Sequence-level detail of second and third sites of the reporter. Artificial piRNAs that associate with MILI and MIWI2 (sequenced at least 1 rpm) are shown. Red triangles indicate MILI slicer activity, while black triangles indicate an unknown endonuclease activity.

(F) The graphs show the 3′-5′end (top) or 5′-5′ end (bottom) distances between secondary piRNAs and downstream inchworm piRNAs on the same genomic strand.

Interestingly, we also observed a second set of 5′ end peaks ∼30 nt downstream (positions +18 to +33) from the 5′ ends of the secondary piRNAs (Figure 1B). The 5′ ends of these reads lie entirely within the artificial LacZ sequence of the reporter and were not designed to be targeted by MILI slicing. Such 5′ ends are generated by an unknown nuclease that is likely recruited by the initial targeting of the reporter by MILI slicing. We call the downstream sequences carrying these 5′ ends “inchworm” piRNAs based on their similarity to piRNAs generated by PIWI slicing in the fly ovarian germline (Homolka et al., 2015). Like secondary piRNAs, the inchworm piRNAs are also loaded into both MILI and MIWI2 (Figure 1B). Inchworm piRNAs in both MILI and MIWI2 display a strong preference for carrying a 5′ end uridine (1U) (Figures 1C, 1D, and S1E). Since secondary piRNAs in MILI do not reveal such a preference, we propose that 1U bias of inchworm piRNAs is likely due to the specificity of the putative nuclease involved in their generation. Nevertheless, as evidenced by the high 1U bias of secondary piRNAs in MIWI2 (Figure 1D), an additional enrichment of particular sequences due to binding affinity of the PIWI protein is also possible (Cora et al., 2014).

Mapping of 3′ ends (Figure 1B) of the reporter-derived reads suggests that 5′ end generation of the inchworm piRNA may be coupled to the 3′ end formation of the preceding secondary piRNA (Figure 1E). The putative nuclease generating these 5′ ends cleaves the reporter ∼37 nt downstream of the 5′ end of the secondary piRNA, which is the most frequent 5′ to 5′ distance between the secondary and inchworm piRNAs (Figure 1F). In addition to this distance constraint, presence of uridines within this range may influence the decision to cleave. Subsequently, the 3′ end of the secondary piRNA is matured by trimming to obtain ∼26 nt MILI or ∼28 nt MIWI2 piRNAs. This corresponds to ∼10 nt trimming, which is seen as the most frequent 3′ to 5′ distance between the piRNAs (Figure 1F). The existence of such a trimming event during production of endogenous piRNAs is supported by the presence of MILI-bound 32–40 nt piRNA intermediates in mice lacking the piRNA biogenesis factor Tdrd2 (Saxe et al., 2013).

Taken together, our mouse reporter reveals how MILI slicing recruits the target transcript as a substrate for piRNA biogenesis. We find that the downstream cleavage fragment is used to generate a series of non-overlapping piRNAs. Although our reporter design restricts unambiguous visualization to only two of these piRNAs, the presence of additional 1U-containing piRNAs not explained by MILI slicing (labeled as “other” in Figure 1C) might mean that processing spreads further in the 5′→3′ direction (but also see Figure 6C).

Figure 6

MIWI2 piRNA Biogenesis by Inchworming and Factors Involved

(A) A scheme showing factors involved in mouse piRNA biogenesis in the embryonic male germline. EXD1 is a new factor identified in this study. See also Figure S6.

(B) Inchworming process in flies and mice. Cytosolic PIWI slicing by MILI or Ago3 leads to loading of fragments from the target as new piRNAs in a nuclear PIWI (Piwi or MIWI2) or cytosolic PIWI proteins (Aub or MILI). In flies, where three PIWI proteins are involved, there is a strict partitioning of secondary and inchworm piRNAs into distinct PIWI proteins. In contrast, in the mouse embryonic male germline, these piRNAs enter both MILI and MIWI2 without any apparent bias.

(C) The plot shows the production of blocks of non-overlapping 1U-containing piRNAs arising from the DsRed coding sequence of the reporter preceding the first MILI cleavage site. Individual piRNAs produced from the reporter are plotted above the vertical dashed line, while the piRNAs targeting the reporter are shown below the line. The piRNAs are in different colors based on the presence of 1U and 10A. Notice that piRNA production continues for several hundred bases. See also Figure S1F.

(D) Calculation of 3′-5′ distances between the piRNAs produced from the DsRed region. A regular spacing with frequent ∼10–20 nt distance is noted.

Exd1 Is a Component of the piRNA Pathway

Apart from MILI slicer, biogenesis of MIWI2 piRNAs is dependent on a number of other factors, including mouse Tudor domain containing 12 (mTDRD12) (Pandey et al., 2013). We identified the uncharacterized Exonuclease domain-containing 1 (Exd1) as a component of Tdrd12 complexes (data not shown). Mouse EXD1 (mEXD1) is composed of 570 amino acids that encode for an N-terminal Like-Sm (Lsm) domain and a central exonuclease domain, followed by a long C-terminal tail (Figure 2A). Using specific antibodies (Figure S2A), we confirm the presence of endogenous mEXD1 in mTDRD12 complexes isolated from mouse testes (Figures 2A and S2B). This interaction is likely to be direct, as we can demonstrate it using tagged proteins expressed in human cell cultures (HEK293T) that are devoid of any piRNA pathway proteins (Figure 2B).

Figure 2

EXD1 is an Interaction Partner of the MIWI2 piRNA Biogenesis Factor TDRD12

(A) Domain architecture of mouse (m) TDRD12 and EXD1 proteins. Immunoprecipitation (IP) of endogenous mTDRD12 from mouse testes lysates and identification of proteins by western blotting. Myc-mExd1 present in transfected mammalian cell culture (HEK293T) lysates serves as a positive control for the mEXD1 antibody. See also Figure S2.

(B) HEK293T cells were co-transfected with tagged protein expression constructs and complex formation tested by anti-HA IP and western blotting.

(C) Domain architecture of Bombyx (Bm) proteins. The HA-BmTdrd12 used has an N-terminal deletion (see reference Pandey et al., 2013) and is considered as full-length throughout in this study.

(D and E) Anti-HA IP and detection of endogenous proteins from transfected BmN4 Bombyx cell cultures.

(F) (Top) IP of endogenous BmTdrd12 from transfected BmN4 cells and detection of partner proteins. Treatment with (+) or without (−) RNases is indicated. (Bottom) Anti-HA IP from co-transfected BmN4 cells and detection of interaction partners. Treatment with RNases is indicated as fold increase in concentration (conc.) of nucleases and time (in minutes) of incubation.

(G) Co-expression of tagged proteins and IP to assess interaction. Myc-BmExd1 interacts with tagged BmTdrd12 and Siwi only.

(H) Endogenous BmExd1 is detected in perinuclear granules called nuage in BmN4 cells and co-localizes with transiently expressed HA-tagged BmTdrd12 or Vasa, both of which are established nuage components. Scale bar is indicated.

To further explore Exd1 in a cellular context with an active piRNA pathway, we used the insect cell line BmN4 derived from Bombyx mori (Silkworm) ovaries (Kawaoka et al., 2009). Similar mammalian systems are currently unavailable. Homology search of a BmN4 transcriptome dataset (Xiol et al., 2014) identified a protein of 315 aa as the Bombyx Exd1 (BmExd1) ortholog. The Bombyx protein still harbors the Lsm and nuclease domains but lacks the long C-terminal tail present in the mouse ortholog (Figure 2C). Using transfected BmN4 cell lysates, we demonstrate that HA-BmTdrd12 associates with endogenous BmExd1 (Figure 2D and Figure S2C) and Myc-tagged BmExd1 (Figure S2D). This association was also confirmed in the reverse direction, as we demonstrate the presence of endogenous BmTdrd12 in HA-BmExd1 complexes (Figure 2E). RNA is not required for this complex formation, as treatment of BmTdrd12 complexes with RNases did not affect retention of BmExd1 (Figure 2F). Furthermore, this complex is an integral component of the piRNA pathway active in the BmN4 cells, as the PIWI protein Siwi is also part of this complex (Figures 2D and 2F). This association with PIWI proteins is likely to be conserved in mice, as MILI is already shown to be part of mTDRD12 complexes in the mouse testes (Pandey et al., 2013).

BmExd1 appears to be exclusive to BmTdrd12 complexes, as we did not find any association between Myc-BmExd1 and HA-tagged versions of various Bombyx piRNA pathway factors: Vreteno (Vret), Papi, Vasa, and Spn-E (Figure 2G). We detected only a weak presence of BmExd1 in HA-Siwi complexes, likely reflecting the fact that not all of Siwi is present in association with BmTdrd12 and BmExd1 (Figure 2G). Note that the second PIWI protein (Ago3) present in the BmN4 system is not present in Tdrd12 complexes (Figures 1D and 1G). Consistent with their biochemical association, both HA-BmTdrd12 and endogenous BmExd1 occupy Vasa-positive cytoplasmic perinuclear granules; identifying BmExd1 as a component of the nuage in BmN4 cells (Figure 2H). Taken together, we identify Exd1 as a conserved interaction partner of the piRNA biogenesis factor Tdrd12 in mammalian and insect systems.

Exd1 Lacks RNase Activity

To understand the molecular role of Exd1, we tested it in nucleases activity assays. Recombinant BmExd1 was incubated with a radiolabelled single-stranded RNA (ssRNA), and RNA cleavage activity was monitored by resolving the reaction products using 10% denaturing urea-PAGE (Figure 3A). Despite screening different buffer conditions, divalent metal ions, various RNA substrates, and different protein constructs (Figure S3A), we are unable to demonstrate RNase activity for BmExd1. Exd1 has a nuclease domain similar to that found in the DEDD superfamily of exonucleases, which are characterized by four conserved acidic residues (D-E-D-D) that coordinate two divalent metal ions essential for catalysis (Zuo and Deutscher, 2001). Examination of a sequence alignment (Figure 3B) indicates that these residues are not conserved in BmExd1 (SDLN) and mEXD1 (AELE) or other orthologs across the animal kingdom, explaining the absence of the nuclease activity.

Figure 3

Exd1 Is Not Active as a Ribonuclease and Exists as Homodimers

(A) Recombinant BmExd1 lacks ribonuclease activity against ssRNA. See also Figure S3. Quality of recombinant proteins used in the experiments is shown.

(B) Sequence alignment of orthologs of Exd1 and the related DEDD family nuclease Rrp6. The catalytic residues (DEDD) in Rrp6 or their absence in Exd1 proteins is indicated.

(C) Cartoon showing region (catalytic domain only) of BmExd1 that was crystallized. Ribbon representation (bottom view) of BmExd1 dimer (red and blue monomers) with the central β sheet (yellow) in the catalytic domain highlighted. The bottom image (side view) has a space-filling model of one of the monomers.

(D) A zoom of the catalytic core of BmExd1 showing the absence of catalytic DEDD residues that are present in the related Rrp6 from budding yeast.

(E) A zoom of the first helix (α1) in the C-terminal dimerization hook from one monomer (red) interacting with the catalytic core of the other monomer (blue). The residues involved are indicated.

(F) Analytical ultra-centrifugation (AUC) analysis of various BmExd1 fragments indicating dimerization in vitro that is mediated via the first helix (α1) of the C-terminal hook.

(G) Self-association of BmExd1 (HA and Myc-tagged versions) in BmN4 cells, suggesting possible dimerization in vivo. This is dependent on the C terminus of the protein.

Exd1 Exists as Homodimers

To gain further understanding of the protein, we crystallized a construct (73–315 aa) of BmExd1 encompassing the nuclease domain (Figures S3B and S3C; Table 1). The native structure was solved to a resolution of 2.4 Å and reveals a perfectly symmetric dimer held together via their C termini (Figure 3C and Figures S3D–S3F). In this configuration, the core nuclease domains are pointing away from each other. Overlay of the BmExd1 nuclease domain with those of active enzymes (such as yeast Rrp6; PDB: 2HBK) from the DEDD family (Midtgaard et al., 2006) reveals the conservation of the overall fold (an RMSD of 2.1 Å over 135 superimposed Cα atoms), but the key residues (DEDD) that provide metal ion coordination and catalytic competence are absent (Figure 3D).

Table 1

Crystallographic Data Collection Statistics for BmExd1

Crystal Parameters	Gd Derivative	Native
Data collection
Wavelength (Å)	0.976	0.976
Space group	P21212	C2
Cell Dimensions a, b, c (Å)	82.8, 135.7, 51.4	122.5, 81.6, 154.8, (β = 111.1°)
Molecules per asymmetric unit	2	5
Resolution (Å) (final shell)	50.0-1.6 (1.65-1.6)	50.0-2.4 (2.5-2.4)
Observed Reflections	396,412	203,740
Unique Reflections	143,744	53,690
Completeness (%) (final shell)	97.2 (96.7)	96.0 (85.8)
Rmeas (%) (final shell)	6.9 (81.7)	14.7 (97.1)
<I/σ(I)> (final shell)	12.0 (1.9)	6.6 (1.3)
Model quality indicators
FOM (centric/acentric)	0.18/0.40
Rcryst (%) (final shell)	15.7 (22.5)	22.1 (32.6)
Rfree (%) (final shell)	18.2 (24.1)	24.7 (31.6)
Mean B-factor (Å2)	25.6	60.4
RMSDs, bonds (Å)/angles (o)	0.03/2.50	0.02/2.0

To examine whether BmExd1 forms dimers in solution, we performed sedimentation velocity measurements using analytical ultracentrifugation (AUC). This indicated that the majority of the recombinant full-length BmExd1 protein (predicted molecular weight = 36.5 kDa) exists as a tight homodimer (measured molecular weight = ∼57 kDa) (Figure 3F). In the BmExd1 structure, the C terminus of each monomer folds into a three-helical (labeled here as α1–3) dimerization hook that reaches across to grab the other monomer at the base of the nuclease core (Figure 3C). Residues from all three helices engage in multiple interactions with conserved residues on the surface of the nuclease core domain (Figure 3E). To determine their relative importance, we prepared a number of deletion mutants and examined their dimerization propensity by AUC experiments. Deletion of the N-terminal Lsm domain did not affect dimerization, suggesting that only the C-terminal dimerization hook is sufficient. Interestingly, the actual contact between the two monomers is established mainly via the first helix (α1) of the dimerization hook, as deletion of the remaining helices had no effect, while deletion of the first helix rendered the protein monomeric (Figure 3F).

Next, we asked whether BmExd1 exists as dimers in vivo by co-expressing HA- and Myc-tagged BmExd1 in BmN4 cells. Consistent with our in vitro experiments, immunoprecipitation of HA-BmExd1 co-precipitates Myc-BmExd1, supporting the possibility that the protein may exist as dimers in vivo. Importantly, deletions at the C terminus that remove all three helices of the dimerization hook abolish this association in vivo, while deletion of the N-terminal Lsm domain did not (Figures 3G and S3G). Furthermore, mEXD1 also shows this self-association property in transfected human HEK293T cells (data not shown). Taken together, we conclude that Exd1 exists as homodimers and is an inactive member of the DEDD nuclease family that has likely taken up new biochemical roles.

Exd1 Functions as an RNA-Binding Protein

To explore whether the inability of BmExd1 to degrade RNAs might translate into a potential for binding RNAs, we carried out in vitro binding studies with full-length recombinant BmExd1 and ssRNA. First, we used UV light to crosslink potential RNA-protein complexes and resolved them by 10% SDS-PAGE. BmExd1 crosslinks to the radioactively labeled ssRNA and becomes visible as a band when exposed to a StoragePhosphor screen (Figure 4A). BSA was used as a control protein that did not bind RNA. We then carried out electrophoretic mobility shift assays (EMSAs) to detect RNA-protein complexes under native conditions. BmExd1 efficiently bound the radioactively labeled ssRNA in a concentration-dependent manner and resulted in the appearance of slow-migrating RNA species (Figure 4B). Incubation of the RNA probe with BSA did not produce any gel-shifts. RNA binding by BmExd1 is sequence independent as six different ssRNAs responded positively in such gel-shift assays (Figure S4A). We quantified this RNA-binding property using isothermal calorimetry (ITC) measurements (Figure 4C). Full-length BmExd1 bound ssRNA with a moderate affinity (Kd = 27.03 μM), and this remains unchanged in the absence of either the N-terminal Lsm domain or that of the C-terminal dimerization hook (Figure 4C). This indicates that the RNA-binding property resides within the inactive nuclease core domain of BmExd1.

Figure 4

Exd1 Acts as an RNA adaptor in a PET Complex

(A) UV-crosslinking of recombinant BmExd1 to radioactively labeled (γ–32P) ssRNA. BSA is used as a negative control.

(B) EMSA with native gels to demonstrate RNA-binding capacity of BmExd1 to ssRNA-1. RNP, ribonucleoprotein complex of BmExd1 and ssRNA. See also Figure S4 where gel-shifts with other ssRNAs are shown, indicating sequence-independent binding of BmExd1.

(C) ITC measurements to quantify RNA-binding affinity (indicated as Kd) of full-length and deletion versions of BmExd1.

(D) Surface charge representation of BmExd1 dimer contoured from −5 kTe−1 (red) to +5 kTe−1 (blue).

(E) Zoom of catalytic core of one monomer (blue) from the BmExd1 (as seen from the bottom) with an ssRNA (green) modeled into it. The C-terminal dimerization hook from the second monomer is also shown (red).

(F) Structure of canonical Lsm domain from a member of the heptameric Lsm core of snRNPs (PDB: 1I5L) and the non-canonical Lsm domain from Edc3 interacting with a helix from Dcp2 (PDB: 4A54). Note the much shorter N-terminal helix in the non-canonical Lsm domain. A modeled Lsm domain of BmExd1 is also shown (lacking an N-terminal helix).

(G) AUC experiment showing that Lsm domain of BmExd1 is monomeric in solution.

(H) Co-immunoprecipitation of full-length BmTdrd12 with various deletion versions of BmExd1 (see details in Figure 3G) using transfected BmN4 cell lysates. The cartoon indicates the Lsm domain of BmExd1 that is required for interaction with BmTdrd12. Also see Figure S4F for data on mouse proteins.

(I) BmExd1 interacts with the helicase core of BmTdrd12. The second tudor domain was already shown to be critical for association with the PIWI protein Siwi in BmN4 cells (see Pandey et al., 2013). Also see Figure S4G.

(J–K) MultiBac co-expression of indicated proteins in Sf21 or Hi5 insect cells and their purification as recombinant proteins is shown.

(J) Purification of recombinant BmExd1 and BmTdrd12 as a complex, indicating direct interaction.

(K) After purification through several chromatography steps, PIWI protein Siwi, BmExd1, and BmTdrd12 remain associated together as a PET complex. Identity of protein bands was confirmed by western blotting and mass spectrometry.

Our efforts to obtain co-crystals with RNA failed, but examination of the BmExd1 dimer structure reveals a positively charged area on the bottom surface (Figure 4D), which along with the inactive catalytic pocket could form the RNA-binding site. In fact, an ssRNA present in the Rrp6-RNA complex structure (PDB: 4OO1) (Wasmuth et al., 2014) can be modeled into the catalytic pocket of the BmExd1 dimer, revealing how the protein might engage RNA substrates (Figures 4E and S4B). Taken together, our studies indicate that the inactive nuclease core domain of BmExd1 functions as an ssRNA-binding module.

Lsm Domain of Exd1 Engages the Helicase Domain of Tdrd12 to Construct a PET Complex

We then examined how Exd1 might interact with Tdrd12 to deliver its RNA cargo. The Lsm domain was originally described in Sm/Lsm proteins that form hetero-heptameric rings (Figure S4C) that bind small nuclear RNAs (snRNAs) involved in pre-mRNA splicing or histone 3′ end processing (Schümperli and Pillai, 2004). The core structural constituent of the canonical Lsm domain is an acutely bent β sheet composed of five anti-parallel strands and an N-terminal helix (Figure 4F). This N-terminal helix is important for interaction with neighboring Lsm domains within the ring. A number of proteins, including Enhancer of decapping 3 (Edc3), harbor a more compact non-canonical Lsm domain with a much shorter N-terminal helix (Figure 4F) (Fromm et al., 2012). Such Lsm domains are monomeric and do not form rings, instead acting as protein-protein interaction modules. Indeed, AUC experiments reveal that recombinant Lsm domain of BmExd1 is also monomeric in solution (Figure 4G). Interestingly, an amphipathic helix from the Decapping enzyme 2 (Dcp2) interacts with the Lsm domain in Edc3 (PDB: 4A54) (Fromm et al., 2012) in a region occupied by the larger N-terminal helix in canonical Lsm domains (Figure 4F). Since we already showed that the Lsm domain of BmExd1 is not required for RNA binding, we wondered whether it might be mediating interactions with Tdrd12.

We took the several N- and C-terminal truncation mutants of Myc-tagged BmExd1 (Figure 3G) and tested their interaction with HA-BmTdrd12 in Bombyx BmN4 cells. Immunoprecipitated HA-BmTdrd12 associated with co-expressed full-length Myc-Exd1 (Figure 4H). This association was still maintained when C-terminal truncations were introduced into BmExd1 (constructs A and B; Figure 4H). In contrast, deletions at the N terminus of BmExd1 that either removes part of, or the entire Lsm domain, abolished this association (constructs C, D and E; Figure 4H). In a reverse experiment, we show that immunoprecipitates of either full-length HA-BmExd1 or that carrying a C-terminal deletion (construct B) isolated from BmN4 cells contain endogenous BmTdrd12 and Siwi (Figure S4D). However, in line with the above result (Figure 4H), the HA-BmExd1 fragment with an N-terminal deletion of the Lsm domain (construct D) failed to co-precipitate the endogenous partners. Finally, we show that the Lsm domain of BmExd1 is directly responsible for this association, as a recombinant version of the domain produced in E. coli can co-precipitate HA-BmTdrd12 from transfected BmN4 cell lysates, while a control protein does not (Figure S4E). We also show that Lsm domain-mediated interaction of Exd1 with Tdrd12 is conserved in mice (Figure S4F).

To identify the region of BmTdrd12 interacting with the Lsm domain of Exd1, we prepared progressive C-terminal truncations. BmTdrd12 is a multi-domain protein composed of a central DEAD box helicase module flanked by tudor domains and followed by a C-terminal CS domain. We previously showed that the second tudor domain is required for association with the Bombyx PIWI protein Siwi, while the CS domain is required for nuage localization in BmN4 cells (Pandey et al., 2013). A HA-tagged version of BmTdrd12 lacking the CS domain and the second tudor domain still associated with endogenous BmExd1 in transfected BmN4 cells (constructs A and B; Figure 4I). However, partial or complete removal of the helicase core (constructs C, D, and E; Figure 4I) abolished this association. Similar results were obtained while monitoring the association of Myc-BmExd1 with these HA-BmTdrd12 constructs (Figure S4G). Importantly, a construct with only the first tudor domain and the intact helicase core (construct F) is fully capable of retaining interaction with Myc-BmExd1 (Figure S4H). These results indicate that the DEAD box helicase core of BmTdrd12 is the contact point for the Lsm domain of BmExd1.

Next, we used the MultiBac system to co-express tagged BmTdrd12 and untagged BmExd1 in Sf21 or Hi5 insect cells (Bieniossek et al., 2012). After multiple chromatography steps (see Supplemental Experimental Procedures), a pure complex essentially containing these two proteins could be purified, demonstrating their direct interaction (Figures 4J and S4I). We confirmed their identities by mass spectrometry and western blotting. Finally, we prepared a further MultiBac co-expression construct where we added untagged Siwi as a third protein. After affinity purification through several chromatography steps, the three proteins were recovered together in the same fraction, indicating that they form a tight complex (Figure 4K). Together with the data from mouse testes and Bombyx BmN4 cells (Figure 2), these data demonstrate the existence of a piRNA processing complex consisting of a PIWI protein, BmExd1, and the scaffolding protein BmTdrd12, which we termed the PET complex.

Loss of Exd1 Leads to Defective Biogenesis of Antisense piRNAs and Activation of Transposons

To explore the in vivo role of the RNA-binding protein EXD1 in the piRNA pathway, we generated null mice by knocking in a triple-stop codon cassette into the Exd1 gene locus (Figures 5A, 5B, and S5A–S5C). Since loss of Tdrd12 has an impact on piRNAs present in the perinatal male germline, we isolated testes from new born pups (P0) of the heterozygous (Exd1) and homozygous (Exd1−/−) genotypes. After deep sequencing of total small RNAs (19–40 nt), reads were mapped to the mouse genome and normalized to the read-depth of the libraries (Figure S5D).

Figure 5

mEXD1 Is Critical for Biogenesis of MIWI2 piRNAs in the Mouse Male Germline

(A) Creation of the Exd1 knockout mouse by insertion of a triple-stop cassette into the coding sequence of the Exd1 genomic locus. See also Figure S5.

(B) Mouse testes from indicated genotypes. The atrophied testes phenotype is only rarely seen in the mutant animals.

(C) Read length profiles obtained by sequencing of total testicular small RNAs (19–40 nt) from P0 animals. Only reads with 24–32 nt length and which are perfectly mapped to multiple sites in the mouse genome are shown.

(D) 24–30 nt reads that map to multiple locations in the genome are depleted in the Exd1−/− mutant. Specifically, the piRNAs annotated as repeat antisense are affected. The classification of the reads based on their presence in MILI or MIWI2 piRNA datasets shows that mainly the abundance of MIWI2 piRNAs is reduced.

(E) The graphs show number of reads mapping to specific repeats. The abundance of repeat antisense, but not sense reads, is reduced in the Exd1−/− mutant, with the largest effect observed for L1Md_F2. Repeats belonging to LINE L1 are shown in red.

(F) Expression of L1ORF1p (red) in the Exd1−/− mutant embryonic (E18.5) testes as revealed by immunofluorescence analysis. DNA (blue) is stained with DAPI.

(G) Expression of LINE1 retrotransposons in the Exd1−/− mutant revealed by western blotting to detect LIORF1p. Post-natal (dpp) ages of donor animals in days are indicated.

(H) Expression of LINE1 retrotransposons transcripts in the Exd1−/− mutant revealed by northern analyses. Note that not all tested mutant samples reveal de-repression of LINE1. Total RNA from Tdrd1−/− was used as a positive control for robust retrotransposon activation that leads to infertility.

(I) IP of MILI or MIWI2 from neonatal pups (P0) and labeling of associated piRNAs by 5′ end labeling. Notice the reproducible reduction (shown with the red arrow) in piRNAs associated with MIWI2. See also Figure S5F.

(J) Immunofluorescence analysis of indicated proteins in the different genetic backgrounds. Compare the clear nuclear staining of MIWI2 in the Exd1 P0 testes to the diffused nucleo-cytoplasmic accumulation of MIWI2 in the Exd1−/− mutant.

(K) The graphs show the mapping of 5′ ends of reads on the consensus L1 transposon sequence allowing maximum 3 mismatches. Total small RNAs and piRNAs associated with MILI and MIWI2 complexes for the different genotypes are compared.

(L) Percentage of reads starting with a U (1U) or having an A at tenth position (A10) is plotted for multi-mapping reads annotated as LINE L1 repeats present in total small RNA libraries. Whereas 1U-bias is almost unchanged, the frequency of A10 is reduced in the Exd1−/− mutant.

(M) A plot indicating 5′ end overlap of reads mapping to both strands of the L1 transposon consensus sequence. A signature peak (black triangle) indicating a distance of 10 nt separating the 5′ ends is suggestive of PIWI slicing, and this is reduced in the Exd1−/− mutant.

Majority of the piRNAs in the perinatal germline are originating from the repetitive parts of the genome, so it was interesting to note that 24–30 nt reads (corresponding to length of piRNAs), which mapped to multiple genomic loci, were reduced in the Exd1 mutant (Exd1−/−) (Figures 5C and 5D). As expected, genome annotations indicate that such reads originate from repeated transposon-related sequences within the genome. Strikingly, sequences with antisense orientation to transposon sequences are depleted in the Exd1 mutant (Figure 5D). Much of this decrease is suffered by small RNAs targeting active retrotransposons like LINE1 and IAP (Figure 5E). Consistently, L1ORF1p, a translation production from the LINE1 transcript, can be detected in Exd1 mutant testes by western blotting and immunofluorescence analyses (Figures 5F–5G and S5E). However, examination of multiple biological samples by northern analyses indicates that derepression of LINE1 transcripts is a low penetrance phenotype (Figure 5H). We report that Exd1 mutant animals display normal fertility and viability (Figure 5B). Taken together, we conclude that mEXD1 is important for biogenesis of a select set of piRNAs targeting retrotransposons in the perinatal male germline.

mEXD1 Is Required for Biogenesis of MIWI2-Bound piRNAs

MILI and MIWI2 are the two PIWI proteins expressed in the perinatal male germline environment. These were immunoprecipitated and associated with small RNAs revealed by 5′ end labeling. In three independent biological replicates, we noticed unchanged presence of MILI piRNAs, but those associating with MIWI2 were dramatically reduced in the Exd1−/− mutant (Figures 5I and S5F). We note that this reduction in MIWI2 piRNAs in the Exd1−/− mutant is distinct from the complete lack of such RNAs in the Tdrd12 mutant (Pandey et al., 2013). Since nuclear entry of MIWI2 is dependent on its association with piRNAs (Aravin et al., 2008), we examined this in P0 or embryonic (E18.5) Exd1 mutant testes. While MIWI2 is still present in the nucleus, a clear redistribution into the cytoplasm can be observed in the Exd1 mutant (Figures 5J and S5G). This is consistent with the fact that MIWI2 piRNAs are only reduced, and not absent, in the Exd1 mutant.

To analyze in detail the composition of the piRNAs remaining in the Exd1 mutant, we prepared libraries of the PIWI-bound small RNA populations. When mapped to the LINE1 consensus, antisense piRNAs are distributed in both PIWI proteins, but with a majority in MIWI2. Strikingly, such antisense reads are dramatically reduced in the MIWI2 libraries prepared from the Exd1 mutant (Figure 5K). MILI-bound piRNAs also show reduced presence of L1 reads. Next, we analyzed sequence signatures to track contribution of MILI slicing to piRNA biogenesis in the Exd1 mutant. When guided by 1U-containing piRNAs, MILI slicing creates secondary piRNAs that carry an adenosine at the tenth position (A10) (Aravin et al., 2008). Examination of total small RNA libraries from control and Exd1 mutant mice reveals a distinct reduction in the A10 signature in the mutant (Figure 5L). Furthermore, the key feature indicative of slicer action, the 10-nt overlap of piRNA 5′ ends, is also reduced in the Exd1 mutant (Figure 5M and Figure S5H). These indicate that formation of the PET complex is required for the MILI slicing-triggered biogenesis/loading of MIWI2 piRNAs.

Discussion

MIWI2 piRNA Biogenesis by Inchworming

We investigated how piRNAs associating with the nuclear PIWI protein MIWI2 are created in the mouse male germline. A wealth of genetic mutant analyses has established a clear hierarchy of mouse piRNA biogenesis factors (Figure 6A), where biogenesis of MIWI2 is dependent on MILI slicer activity (Aravin et al., 2008, De Fazio et al., 2011, Kuramochi-Miyagawa et al., 2008). So it was always puzzling as to why the majority of MIWI2 piRNAs lack the sequence signature consistent with an origin via MILI slicing (i.e, robust 10 nt overlap of its 5′ ends with that of MILI piRNAs) (Aravin et al., 2008). Instead, MIWI2 piRNAs display an exceptionally high 1U-bias, a feature linked to primary processing.

Here we demonstrated that MILI slicing licenses a target transcript to enter the piRNA biogenesis pathway to generate a series of 1U-containing piRNAs in the 5′→3′ direction, a process we previously termed in flies as inchworming (Homolka et al., 2015). Such piRNAs associate with both MILI and MIWI2 (Figure 6). However, since MIWI2 piRNA levels are disproportionately affected in the MILI-slicer mutant (De Fazio et al., 2011), we believe that this process primarily serves to populate MIWI2 with piRNAs. Curiously, the region upstream (∼1 kb DsRed coding sequence) of the piRNA binding sites in the reporter also entered the piRNA pathway (Figures 6C and S1F). The generated piRNAs carry a 1U bias and arise in a non-overlapping manner with regular spacing between them (Figures 6C and 6D). We are unable to determine whether a piRNA binding at the 5′ end is the trigger for this processing event. So we propose that although inchworming proceeds essentially with 5′→3′ directionality, the upstream cleavage fragment generated by MILI slicing might also be presented to the biogenesis machinery.

Inchworming represents a very efficient mechanism by which germ cells can extract a large amount of sequence information in the form of small RNAs from the targets of PIWI slicing. In the mouse perinatal male germline, piRNAs guiding the MILI endonuclease display an overall sense orientation to L1 retrotransposon elements (Aravin et al., 2008) (Figure 5K). This ensures that its target sequences are essentially antisense to L1. Thus, inchworming on these targets contributes to amplifying the repertoire of antisense piRNAs available to target L1. Since both MILI and MIWI2 are recipients of the inchworming process, the same piRNAs can be used by MILI to silence transposon transcripts in the cytoplasm, while MIWI2 will repress transposon genomic loci in the nucleus. Supporting this, a mutation that abolishes MILI slicer activity reduces levels of antisense L1 piRNAs and impacts loading of MIWI2 (De Fazio et al., 2011).

A similar mechanism that links cytoplasmic PIWI slicing to nuclear piRNA biogenesis is in operation in the fly ovarian germline (Han et al., 2015, Homolka et al., 2015, Mohn et al., 2015). Pointing to a mechanistic similarity, orthologous factors are involved in loading the nuclear PIWI protein in the fly and mouse systems (Figure 6A). However, despite being conceptually related, the fly and mouse inchworming processes differ significantly. First, in the 3-PIWI system of the fly ovarian germline, cytosolic Ago3 slices the target to load the secondary piRNA into Aubergine, while all downstream inchworm piRNAs are primarily reserved for nuclear Piwi (Figure 6B). In contrast, the 2-PIWI system of the mouse testicular germline uses the generated secondary and inchworm piRNAs to load both MILI and MIWI2 with no apparent specificity (Figure 1). Second, the spacing between consecutive piRNAs is longer in mice (∼37 nt between 5′ ends), requiring extensive trimming to mature the 3′ ends.

Is inchworming similar to primary processing? To address this question one can examine piRNAs present in MILI in the pre-meiotic germ cells of the post-natal (P11) male germline, where it is the only PIWI protein present, and its slicer activity is dispensable for piRNA biogenesis (Di Giacomo et al., 2013). Analysis of such MILI-bound sequences indicates that endogenous primary piRNAs are produced in a phased manner (Han et al., 2015, Mohn et al., 2015), with a spacing that is similar to what we report here for MILI slicer-triggered inchworming in the perinatal germline (Figure 1F). This suggests that although the mechanisms for initiation (PTS or PIWI slicing) of primary processing may vary, subsequent processing of the RNA substrate is the same, likely reflecting the architectural relatedness of the processing machineries involved (Figure 6B).

Exd1 as an RNA Adaptor in the PET Complex for MIWI2 piRNA Biogenesis

Biogenesis of MIWI2 piRNAs is dependent on a distinct set of factors (Figure 6A). In this study, we identified Exd1 as a partner of Tdrd12 in both mouse and insect systems. Our data indicate that Exd1 acts as an RNA adaptor within a PET complex (Figure 4). Given the interaction of BmExd1 with the helicase domain of BmTdrd12 (Figure 4I), it is possible that Exd1 might deliver its RNA cargo directly to the RNA helicase. Our Bombyx Exd1 structures reveal conformational flexibility of the protein dimer. In the “open” state the two monomers are splayed out, revealing a large groove in between (Figures 3C and S3D). In contrast, in the “closed” state one of the monomers displays a 55° rotation (equivalent to >30Å movement in space), narrowing the space between the monomers (Figure S3E). We speculate that such conformational changes by the Exd1 dimer might play a role in delivering its RNA cargo directly to the helicase domain of Tdrd12. Nevertheless, the exact details of how mEXD1 contributes to biogenesis of MIWI2 piRNAs remains to be determined.

Finally, the role of Exd1 may not be conserved in flies. We examined the Exd1 ortholog in Drosophila (CG11263; DmExd1) (Figures S6A and S6B) and report a lack of transposon activation in either knockdown or knockout fly lines (Figures S6C–S6E). Such mutant female flies display normal fertility. Exd1 appears to be a fast-evolving protein, with an expansion of the C-terminal region from insects to human. Strikingly, the Drosophilid genomes carry a reduced version of the protein that seems to lack the Lsm domain needed for interaction with Tdrd12 (Figure S6A). In this context, it is worth mentioning that while Tdrd12 is represented by a single gene in all other organisms, Drosophila melanogaster has an expanded set of three Tdrd12 proteins that are critical for piRNA biogenesis (Handler et al., 2011). Perhaps this is an illustration of an ongoing optimization of transposon defense strategies in individual genomes.

Experimental Procedures

Mouse and Fly Strains

Mouse knockin line (Rosa26-pi) expressing a DsRed reporter from the ubiquitous Rosa26 locus was prepared. The reporter has a 3′-UTR containing perfectly complementary binding sites for 35 different MILI-bound piRNAs present in perinatal mouse testes. The Exd1 knockout line was prepared by inserting a triple-stop codon into the exon 8 of the mouse Exd1 genomic locus. This results in premature termination of the coding sequence and truncation of the protein and/or nonsense-mediated decay. A fly line with minos insertion into the DmExd1 (CG11263) gene locus was obtained from Bloomington Drosophila Stock Center (BL#44902). RNAi hairpin knockdown lines were obtained from VDRC, Austria or from NIG, Japan. See Supplemental Experimental Procedures for further details.

Clones and Constructs

Exd1 cDNAs were isolated from mouse testes, Bombyx BmN4 cell line, and Drosophila ovaries by RT-PCR. A detailed description of all clones and constructs can be found in the Supplemental Experimental Procedures.

Antibodies

Rabbit polyclonal antibodies recognizing endogenous Exd1 proteins were prepared against the following antigens produced in E.coli: Bombyx Exd1 (1–315 aa), Drosophila Exd1 (1–56 aa), and Mouse EXD1 (N-term antigen; 1–126 aa and C-term antigen; 420–570 aa).

Crystallization of BmExd1

Recombinant BmExd1 (73–315 aa) consisting of the catalytic domain and the C-terminal dimerization hook was produced in E.coli. Data collection parameters and crystal statistics are provided in the Supplemental Information.

Small RNA Libraries and Bioinformatics

Mouse PIWI complexes were isolated by immunoprecipitation from testes of neonates (P0) and libraries of associated piRNAs were prepared for deep sequencing. When indicated, total small RNAs (19–40 nt) were used for library preparation. Details of library preparation and analyses are provided in the Supplemental Experimental Procedures.

Author Contributions

Z.Y. identified Exd1 and carried out most experiments in Figures 2, 4, and 5. K.-M.C. performed all structural and biophysical experiments in Figures 3 and 4. R.R.P. supervised mouse breeding and prepared deep sequencing libraries from mouse Piwi immunoprecipitations. D.H. designed the MILI slicer reporter mouse and analyzed all sequencing data. M.R. carried out mouse EXD1-TDRD12 interaction studies. B.K.R.J. produced antibodies to fly Exd1. R.S. aligned mouse sequencing data to the genome. M.-O.F. performed fly crosses and provided expert advice to Z.Y and M.R. for fly experiments. A.A.M. solved the Exd1 structure. R.S.P. coordinated the study and wrote the article with inputs from others.