A conserved PIWI silencing complex detects piRNA-target engagement

Summary

In animal germ cells, PIWI proteins use piRNAs to detect active selfish genetic elements. Base pairing to a piRNA defines transposon recognition, but how this interaction triggers a defensive response remains unclear. Here, we identify a transposon recognition complex composed of the silkworm proteins Siwi, GTSF1, and Maelstrom. Biochemical and cryo-EM analyses show that extended piRNA-target pairing locks Siwi in a conformation that recruits GTSF1 and Maelstrom. Extended piRNA-target pairing is recognized by the N-terminal helix of Maelstrom and the first zinc finger of GTSF1, which act together to hold Siwi in an endonucleolytically active state. The resulting activated complex, termed Siwi*, rapidly cleaves target RNAs and recruits the piRNA biogenesis factor Spindle-E. Structural predictions reveal related complexes in animals ranging from humans to sponges, indicating PIWI* assembly is a conserved transposon recognition mechanism employed broadly across the metazoan kingdom.

eTOC Blurb

De et al. show how PIWI proteins, which protect the genome by recognizing and silencing transposons, assemble into a silencing complex only upon binding extended RNA targets. The activated complex, termed PIWI*, is conserved and functions as a molecular checkpoint to ensure accurate piRNA-guided genome defense in animal germ cells.

Graphical Abstract

Introduction

Recognizing and responding to biological threats is fundamental to all life. Accordingly, animals have evolved an array of molecular defense systems to detect and neutralize potentially harmful entities^1,2. Pathways that sense and react to foreign molecules associated with bacterial and viral infections are well-characterized, with many components understood in atomic detail³. By contrast, the mechanisms by which animals detect resident threats, such as active transposable elements and recently acquired retroviruses, are less understood.

The piRNA pathway is an essential defense system that protects the metazoan germline from transposons and retroviruses⁴. piRNAs are genome-encoded small RNAs (~28 nt) that often exhibit sequence complementarity to transposon-derived transcripts^5,6. piRNAs associate with PIWI proteins^7–11, which facilitate base-pairing interactions with target RNAs that represent potential genomic threats^12,13. In the nucleus, piRNA-target pairing triggers transcriptional silencing and heterochromatin formation at target loci^14–17. In the cytoplasm, piRNA-target pairing promotes PIWI-catalyzed cleavage of target RNAs, which are then fragmented to generate new piRNAs, amplifying piRNA sequences associated with successful target recognition^18–21.

While base pairing to a piRNA defines target recognition, how this interaction triggers a defensive response remains unclear. In flies, tethering Piwi to a nascent transcript fails to induce transcriptional silencing, suggesting that additional factors are required to detect piRNA-target engagement^22–25. Likewise, piRNA amplification follows a structured process, where cleaved target RNAs are faithfully transferred to new PIWI proteins, implying that target recognition and cleavage are linked to downstream processing^18–21. Although genetic and biochemical studies have identified hundreds of proteins involved in piRNA-mediated transposon silencing, the key factors that directly interact with PIWI proteins to translate target recognition into a silencing signal remain unknown.

Results

GTSF1 and Maelstrom detect extended piRNA-target pairing

To identify proteins that directly interact with PIWI proteins, we applied 3D structure prediction using an AlphaFold-based pipeline^26–28. Previous RNAi screens implicated 274 Drosophila ovary proteins in transposon silencing^29–31. We predicted the structure of each protein in complex with Drosophila Piwi, Aubergine (Aub), and Argonaute3 (Ago3). Of these, only seven produced high-confidence interchain interactions with at least one Piwi protein (Fig. 1A): four GTSF1 paralogs (Arx, CG34283, CG32625, and CG14036), Maelstrom (Mael), Vreteno (Vret), and Hen1. These predictions are supported by previous results showing Drosophila Piwi co-immunoprecipitates with Arx^32,33, Mael²⁵, Vret³⁴, and Hen1³⁵. GTSF1 proteins have also been shown to co-purify with PIWI proteins from mice³⁶ and silkworms³⁷, and activate Piwi-catalyzed target RNA cleavage across species^37,38. Extending our analysis to core piRNA factors in silkworm, hydra, sponge, mouse, and human revealed that GTSF1 and Mael are consistently predicted as strong PIWI interactors across evolutionarily distant animal species (Fig. 1B).

Fig. 1. GTSF1 and Mael recognize piRNA-target pairing.

(A) Interchain predicted aligned error (iPAE) scores for Piwi, Aub, or Ago3 in complex with each of 274 Drosophila ovary proteins implicated transposon silencing. Inset shows complexes with non-zero scores. Dark blue indicates the highest confidence interchain predictions. (B) iPAE scores for predicted structures of Piwi proteins from Bombyx mori (silkworm), Hydra vulgaris (hydra), Amphimedon queenslandica (sponge), Mus musculus (mouse), and Homo sapiens (human) with core piRNA proteins from the same animal. (C) Particle distribution measured by mass photometry for complexes involving the Siwi-piRNA complex with silkworm GTSF1, Mael, and target RNA. GTSF1 was fused to maltose-binding protein (~50 kDa total) to enable detection by mass photometry, which cannot reliably measure proteins under 30 kDa. Numbers describing target RNAs indicate the range of nucleotides with complementarity to the piRNA bound to Siwi (counting from the piRNA 5’ end). Also see Fig. S1.

To characterize these interactions, we performed biochemical reconstitution using purified recombinant GTSF1, Mael, and Siwi from the silkworm Bombyx mori, leveraging our previously established system for producing Siwi-piRNA complexes¹³. Mass photometry analysis failed to detect an interaction between GTSF1 and the Siwi-piRNA complex in the absence of target RNA (Fig. 1C and S1). However, upon addition of a target RNA, a peak corresponding to the Siwi-piRNA-target-GTSF1 complex emerged, indicating that GTSF1 recruitment depends on target engagement. This recruitment was strongly influenced by base-pairing to the 3’ end of the piRNA: target RNAs with complementarity limited to nucleotides 2–19 or 2–21 failed to support complex formation with GTSF1. Similarly, a small peak corresponding to the Siwi-piRNA-target-Mael complex was observed only when the target RNA contained extensive 3’ pairing. When all components were combined, we observed a peak consistent with the Siwi-piRNA-target-GTSF1-Mael pentameric complex, but only in the presence of target RNA. These results demonstrate that GTSF1 and Mael are selectively recruited to Siwi upon target engagement, suggesting that these factors function as conserved sensors of piRNA-target pairing. Borrowing from the spliceosome field, we term the pentameric Siwi-piRNA-target-GTSF1-Mael complex Siwi* as structural and mechanistic analyses indicate this is an activated form of Siwi (see below).

Siwi adopts a locked, pre-catalytic conformation upon piRNA-target pairing

We next examined the reconstituted complexes by cryo-EM and single-particle analysis, starting with Siwi-piRNA, GTSF1, and a target complementary to piRNA nucleotides 2–25 (Fig. 2A, S2–S3, Methods S1, and Table 1). Surprisingly, apart from a four-residue peptide (residues 101–104) bound to the bottom of the Siwi PIWI domain, we detected no density corresponding to GTSF1. This finding is reminiscent of previous reconstructions of GTSF1 with Ephydatia fluviatilis Piwi (EfPiwi), where GTSF1 also remained unresolved, supporting a model in which PIWI-GTSF1 interactions are dynamic³⁹. Despite the absence of visible GTSF1, this structure provided a valuable view of Siwi engaged with a piRNA-target duplex. As recently shown for Mili and EfPiwi³⁹, upon extended piRNA-target pairing, Siwi undergoes a rearrangement involving the release of the piRNA 3’ end and movement of the PAZ and N domains into a “locked” conformation (Fig. S10).

Fig. 2. Structural basis for piRNA-target recognition.

(A). cryo-EM density and cartoon representation of the Siwi-piRNA-target-GTSF1 complex. Siwi is colored gray, with major domains labeled. Minor density corresponding to GTSF1 (orange) was observed at the bottom of the PIWI domain. Top inset shows density corresponding to the piRNA-target duplex extending out of the Siwi central RNA-binding cleft. Duplex positions, counting from the piRNA 5’ end, are labeled. (B) Cryo-EM density and model of the Siwi* complex. Top inset shows close-up of the Mael MAEL domain engaging the Siwi MID and PAZ domains. Bottom inset shows GTSF1 interacting with the PIWI, L1, and N domains. (C). Linear schematics of Siwi* proteins and their domains. (D) Close-up views of contacts to piRNA-target duplex made by GTSF1 (top) and Mael (bottom). Target cleavage was inhibited by the omission of exogenous Mg^2+,40. Also see Figs. S2–S5.

Table 1.

Cryo-EM data collection, refinement, and validation statistics

Sample	Siwi-piRNA-target-GTSF1	Siwi-piRNA-target-GTSF1-Maelstrom (Siwi*)	Siwi-piRNA-target-GTSF1-Maelstrom (Siwi* conf. 2)	Siwi-piRNA-target-Maelstrom	Siwi-piRNA-target-GTSF1-Maelstrom-Spindle-E (Siwi*-Spn-E)
Deposition ID	(EMDB-49407) (PDB 9NHB)	(EMDB-49408) (PDB 9NHC)	(EMDB-49409) (PDB 9NHD)	(EMDB-49410) (PDB 9NHE)	(EMDB-49422) (PDB 9NHS)
Data collection
Microscope	Talos Arctica	Talos Arctica	Talos Arctica	Talos Arctica	Talos Arctica
Camera	Gatan K2 Summit	Gatan K2 Summit	Gatan K2 Summit	Gatan K2 Summit	Gatan K2 Summit
Camera mode	counting	counting	counting	counting	counting
Magnification	45,000	45,000	45,000	45,000	45,000
Voltage (kV)	200	200	200	200	200
Total electron exposure (e−/Å2)	50	50	50	50	50
Exposure rate (e−/pixel/s)	8.9	9.5	8.0	8.0	9.2
Exposure time (s)	4.6	4.4	5.2	5.2	4.5
Frame rate (ms)	200	200	200	200	200
Number of frames	23	22	26	26	23
Defocus range (μm)	−1.0 to −3.0	−1.0 to −3.0	−1.0 to −3.0	−1.0 to −3.0	−1.0 to −3.0
Pixel size (Å)	0.91	0.91	0.91	0.91	0.91
Symmetry imposed	C1	C1	C1	C1	C1
Data acquisition software	Leginon	Leginon	Leginon	Leginon	Leginon
Micrographs collected (no.)	4687	5281	4802	4991	3412
Data analysis
Total extracted particles (no.)	2,411,587	1,470,893	2,346,312	2,268,300	611,734
Particles used for 3D (no.)	1,505,526	1,105,389	795,224	397,459	530,244
Final refined particles (no.)	161,717	198,296	90,700	158,398	83,871
Symmetry	C1	C1	C1	C1	C1
Global resolution (Å)
FSC 0.5 (unmasked / masked)	6.3/4.1	4.8/4.0	7.1/4.5	6.3/4.5	7.9/5.8
FSC 0.143 (unmasked / masked)	4.3/3.7	4.2/3.6	4.6/4.1	4.5/4.2	4.9/4.2
Local resolution range (Å)	3.1–7.1	3.0–5.9	3.4–8.9	3.4–7.7	4.0–9.0
3DFSC Sphericity (%)	90.4	96.4	84.5	84.3	78.2
Map sharpening B-factor (Å2)	161.4	173.4	161.7	190.9	127.4
Model composition
Chain	5	6	6	5	7
Non-hydrogen atoms	7060	10438	9242	8785	18026
Protein residues	752	1155	1017	972	2117
Nucleotides	46	52	48	44	52
Ligands	Mg: 1	5 (Mg: 2, Zn: 3)	Mg: 1	Mg: 1	4 (Mg: 1, Zn: 3)
Model Refinement
Refinement Package	Phenix/ISOLDE	Phenix/ISOLDE	Phenix/ISOLDE	Phenix/ISOLDE	Phenix/ISOLDE
MapCC (volume/mask)	0.77/0.79	0.81/0.84	0.74/0.74	0.67/0.68	0.68/0.69
B-factors (Å2)
Protein residues	51.99	52.09	119.17	117.32	176.89
Nucleotides	63.50	50.84	122.27	116.35	161.54
Ligands	44.02	31.39	97.89	97.89	60.10
R.m.s. deviations
Bond lengths (Å)	0.010	0.006	0.008	0.009	0.007
Bond angles (°)	1.650	1.140	1.365	1.626	1.293
Validation
Map-to-model FSC 0.5	3.9 / 3.8	3.8/3.7	4.4/4.3	4.5/4.4	5.4/4.5
(unmasked/masked) Map-to-model FSC 0.143 (unmasked/masked)	3.6 / 3.6	3.6/3.5	4.1/4.0	4.1/4.1	4.2/4.2
Ramachandran plot
Favored (%)	95.82	95.35	93.57	93.82	93.85
Allowed (%)	4.18	4.65	6.43	6.18	6.15
Outliers (%)	0.00	0.00	0.00	0.00	0.00
MolProbity score	1.51	1.63	1.92	1.89	2.04
Clashscore	4.38	5.68	9.36	8.84	13.11
Poor rotamers (%)	0.30	0.00	0.11	0.34	0.48
CaBLAM outliers (%)	1.78	2.05	2.47	2.35	2.50
EMRinger score	3.42	3.22	1.66	1.17	0.71

We noticed two key features of the locked Siwi structure. First, the observed Siwi conformation is catalytically inactive because E708, a cleavage-essential residue⁴⁰, remains “unplugged” from the active site (Fig. S4). Second, Siwi does not contact RNA beyond position 19 of the piRNA-target duplex (Fig. 2A). This was surprising because silkworm piRNAs are 25–29 nucleotides in length⁴¹. Thus, Siwi alone cannot fully interrogate the pairing status of its piRNA guide.

GTSF1 and Mael recognize the Locked Siwi Conformation

We next examined the full Siwi* complex by cryo-EM (Fig. 2B, S2–S3, and Table 1). The reconstruction contains density for most of GTSF1, which forms a ring around the MID–PIWI lobe of Siwi, with the GTSF1 N-terminal zinc finger (ZnF) domains contacting the PIWI domain and a small C-terminal domain (CTD) composed of a strand-turn-helix motif that binds at the N–L1–PIWI interface. Mutating Siwi-contacting residues in the ZnF1 or deleting the CTD weakened the affinity of GTSF1 for Siwi but did not abolish the ability to activate target cleavage (Fig. S5). Mael sits atop GTSF1, bridging the Siwi central cleft via interactions with the MID and PAZ domains.

Importantly, both GTSF1 and Mael specifically recognize the locked conformation of Siwi (Fig. S4). The GTSF1 CTD docks into a groove formed by the PIWI, L1, and N domains, which is a unique feature of the locked state. Mael engages the rotated PAZ domain in a similarly locked conformation-specific arrangement (Fig. 2B). Thus, as originally proposed for GTSF1 and EfPiwi by Li and coworkers³⁹, silkworm GTSF1 and Mael recognize structural features unique to the target-engaged conformation of Siwi.

GTSF1 and Mael create an extended piRNA-target recognition surface

In addition to contacting Siwi, GTSF1 and Mael directly bind the piRNA-target duplex (Fig. 2B). The ZnF1 domain of GTSF1 engages target nucleotides at positions 21–23, straddling the adjacent major and minor grooves (Fig. 2D). Mutation of the corresponding Znf1 residues in sponge and mouse GTSF1 proteins weakened activation of Piwi-catalyzed cleavage^38,39. Additionally, the GTSF1 CTD interacts with piRNA nucleotides 16–17 and the duplex major groove around position 20. Mael contains a long N-terminal α-helix that reaches out, like a long arm, from the central MAEL domain to the 3’ end of the piRNA (Fig. 2B). This “arm-helix” positions an ordered loop that docks with the target duplex at positions 23–25 (Fig. 2D). Mael also has an N-terminal High Mobility Group (HMG), which is disordered in our reconstruction but may also contribute to interactions with the end of piRNA-target duplex. The combined contacts explain why GTSF1 and Mael are specifically recruited upon piRNA 3’ pairing (Fig. 1C). Moreover, the structure shows that GTSF1 and Mael collaborate with Siwi to create an extended piRNA-target-binding surface, enabling the detection of base-pairing along the full length of the duplex.

GTSF1 and Mael shape target cleavage preferences

Because GTSF1 and Mael interact with the 3’ end of the piRNA-target duplex (Fig. 3A), we examined their role in target recognition using an RNA Cleave-’n-Seq assay⁴². We generated a library of 1024 target RNAs, each complementary to piRNA nucleotides 2–14 and either matched or mismatched to nucleotides 16–25 (Fig. 3F). The library was incubated with Siwi-piRNA alone, with Siwi-piRNA and GTSF1, or with Siwi-piRNA, GTSF1, and Mael. Bulk cleavage analysis showed that GTSF1 increased both the overall cleavage rate and the fraction of targets cleaved (Fig. S5).

Fig. 3. GTSF1 and Mael expand piRNA-target recognition.

(A). Surface representations illustrating piRNA-target contacts made by Siwi, Siwi-GTSF1, and Siwi-GTSF1-Mael. (B). Bee swarm plots display the relative abundances of 1024 target RNAs (compared to a no-cleavage control) for libraries incubated with 10 nM Siwi (left), 10 nM Siwi + 10 nM GTSF1 (middle), or 10 nM Siwi + 10 nM GTSF1 + 250 nM Mael (right) for 15 minutes at 27 °C. Data were fit to a normal distribution, and cleaved targets (gray circles) were defined as sequences depleted by ≥3σ. Orange and purple circles indicate target sequences most influenced by the addition of GTSF1 and Mael, respectively. (C). Bar graph shows the average frequency of piRNA-complementarity at each randomized position across all target RNAs cleaved by Siwi alone. (D-E). Bar graphs showing average piRNA-complementarity across targets cleaved by Siwi + GTSF1 (D) or Siwi + GTSF1 + Mael (E). (F) Schematics of the library design: constant region complementary to the 5’ side of the piRNA guide and the randomized region (match or mismatch) to the ten 3’-terminal nucleotides. (G-H). Bar graphs of piRNA-complementarity frequencies across targets most affected by GTSF1 (G) and Mael (H). n indicates the number of cleaved target sequences in each data set. Also see Fig. S5.

To assess sequence preferences, we performed RNAseq on cleavage reactions, using depletion of each sequence relative to a no-cleavage control as a measure of cleavage efficiency (Fig. 3B and Methods S1). Targe cleavage by Siwi alone exhibited a strong preference for pairing at positions 16–18, which gradually diminished toward the 3’ end, with a notable exception at position 22 (Fig. 3C). The addition of GTSF1 increased the overall degree of depletion and shifted sequence preferences (Fig. 3B): with GTSF1, the average dependence on pairing at positions 16–18 was less, while sensitivity to pairing at positions 22–23 was high (Fig. 3D). Targets most affected by GTSF1 displayed an increased bias for pairing at positions 21–23, primary contact points with its ZnF1 domain (Fig. 2D).

Including Mael further enhanced cleavage efficiency and expanded the range of cleaved sequences (Fig. 3B). Mael further reduced dependence on pairing at positions 16–19 and lessened the bias for pairing at 22–23 (Fig. 3E). A similar effect was found using high concentrations of GTSF1, suggesting that the loss of Mael can be largely compensated for by an excess of GTSF1 (Methods S1). Mael targets most affected by Mael display a strong preference for pairing at positions 23–25 (Fig. 3H). Thus, by engaging the extended piRNA-duplex, GTSF1 and Mael reshape target recognition, reducing Siwi’s intrinsic reliance on early 3’ pairing while enhancing extended 3’ pairing sensitivity. This broadens the spectrum of cleavable targets, allowing piRNAs to recognize evolving genomic threats without a strong requirement for complementarity at any single position.

GTSF1 Znf1 stabilizes the plugged-in Siwi active site

To investigate how GTSF1 and Mael activate Siwi-mediated cleavage, we compared the structures of Siwi-piRNA-target-GTSF1 and the fully assembled Siwi* complex. The Siwi Cα backbone remains largely unchanged between the two, except for two flexible loops in the PIWI domain (Fig. 4A). One of these loops, PIWI loop 2, contains the catalytic residue E708. In the Siwi-piRNA-target-GTSF1 structure, loop 2 is partially disordered, leaving E708 in an unplugged, inactive conformation, removed from the active site. By contrast, in the Siwi* complex, loop 2 is fully ordered, and E708 is shifted into a “plugged-in” conformation, resembling the catalytic state of T. thermophilus Argonaute^43,44. This conformational shift appears to be driven by the ZnF1 domain of GTSF1, which occupies the space normally taken by loop 2 in its unplugged configuration (Fig. 4B). A cryo-EM reconstruction of the Siwi-piRNA-target-Mael complex (lacking GTSF1) confirms this model, as it shows loop 2 remains unplugged in the absence of GTSF1. Further supporting this model, we identified a lower-resolution Siwi* reconstruction (conformation 2) in which the ZnF domains were disordered, and loop 2 was again unplugged. Thus, ZnF1 of GTSF1 drives Siwi activation by sterically preventing loop 2 from adopting the unplugged conformation.

Fig. 4. Structural basis for Siwi activation by GTSF1 and Maelstrom.

(A). Superposition of Siwi Ca atoms from Siwi-GTSF1 and Siwi* models. Inset focuses on the two mobile PIWI domain loops. (B). Top: Side-by-side comparisons of cryo-EM reconstruction of Siwi-piRNA-target-GTSF1, Siwi*, Siwi* conformation 2 (Fig. S4), and Siwi-piRNA-target-Mael. Siwi proteins are colored gray, GTSF1 orange, and Male purple. Bottom: Close-up views of PIWI loop 2 in models built from each reconstruction. Wire mesh depicts experimental density. Catalytic residue E708 adopts the plugged-in conformation in the Siwi* structure. (C). RNA Cleave-n’-seq data for target depletion by 10 nM Siwi (left) and 10 nM Siwi + 250 nM Mael (right) after 15 minutes at 27 °C. (D). Fraction of target RNA library (1 nM) cleaved by 10 nM Siwi alone, or 10 nM Siwi with 250 nM Mael, or 10 nM Siwi with 250 nM Mael-ΔN, which has 88 residues removed from the N-terminus (including the HMG domain and about half of the arm-helix), over time. Also see Fig. S2–S4.

We also identified a structural link between the ZnF1 domain of GTSF1 and the N-terminal arm-helix of Mael. In the fully assembled Siwi* structure, these two elements contact each other, suggesting a functional relationship (Fig. 2B). Indeed, in the Siwi* reconstruction where ZnF1 is disordered, the Mael arm-helix is also absent. Similarly, the arm-helix is missing in the Siwi-piRNA-target-Mael reconstruction, in which GTSF1 was omitted from the sample. These observations suggest synergism between ZnF1 and the Mael arm-helix, where each element helps position the other. Thus, when the Mael arm-helix docks onto 3′ pairing, it also promotes ZnF1 positioning, which in turn drives the activation of Siwi-mediated cleavage. This connection provides a physical link between target recognition at the piRNA 3’ end and activation of the Siwi catalytic site.

Maelstrom is a conditional inhibitor of Siwi catalysis

We also examined how Mael impacts target cleavage by Siwi in the absence of GTSF1. To our surprise, Mael alone inhibited the slow basal cleavage activity of Siwi (Fig. 4C). The biological significance of this inhibition is unclear, but its mechanistic basis appears to lie in the N-terminal RNA-binding region of Mael. A truncated Mael protein with 88 residues deleted from the N-terminus, removing the HMG domain and about half of the arm-helix, did not inhibit basal Siwi cleavage (Fig. 4D). In the Siwi–piRNA–target–Mael cryo-EM reconstruction, the arm-helix is not visible, but there is an unexplained accumulation of diffuse density near the 3′ end of the piRNA that might correspond to the end of the arm (Fig. 4B). We propose that, in the absence of GTSF1, the N-terminal region is mobile but can still bind the end of the piRNA-duplex, disrupting alignment within the Siwi active site and inhibiting cleavage. By contrast, in the presence of GTSF1, the arm-helix can dock against the ZnF1 domain, which helps align the RNA for catalysis. Mael is thereby a conditional inhibitor of target cleavage by Siwi.

Siwi* recruits downstream silencing factors

Our data show that GTSF1 and Mael assemble onto Siwi to form Siwi*, which exhibits enhanced target recognition and cleavage activity. Siwi* assembly also creates new inter-protein surfaces that might recruit additional factors for downstream silencing. To investigate this possibility, we used structure prediction to identify potential binding partners of various Piwi* complexes. The sequences of Drosophila Piwi proteins were combined with their putative GTSF1 partners and Mael and screened against the 274 known piRNA factors^29–31. Each complex yielded roughly a dozen proteins with positive interchain PAE scores and two high-confidence hits: Nxf2, a subunit of the SNiFX complex that directs transcriptional silencing^24,45–47, and Spindle-E (Spn-E), a putative DEAH helicase necessary for piRNA amplification via the ping-pong cycle⁴⁸ (Fig. 5A). Although Bombyx mori lacks a clear Nxf2 homolog, Spn-E emerged as the strongest candidate in a screen of Siwi* against core silkworm piRNA factors. Likewise, Tdrd9, the mouse homolog of Spn-E, scored highly in screens involving Miwi* and Mili*.

Fig. 5. Spindle-E is recruited to Siwi*.

(A). Predicted complexes with positive iPAE scores for Drosophila Piwi*, Aub*, and Ago3* with ovary piRNA factors, and all scores for Siwi*, Miwi*, and Mili* predicted with core piRNA factors. Scores reflect the confidence of placing residues in the additional factor relative to all subunits of the respective PIWI* complex. Scores for factors with each Piwi protein alone included for comparison. (B). Particle distribution measured by mass photometry for complexes involving the Siwi, GTSF1, Mael, target RNA, and Spn-E. Spn-E-ΔN is a truncated form of Spn-E with the N-terminal region (32 residues) removed. (C). cryo-EM reconstruction of the Siwi*-Spn-E complex. (D). Ribbon representation of the Siwi*-Spn-E model highlights the position of Spn-E relative to the target RNA. (E). Close-up view showing interactions between the Spn-E N-terminal region and Siwi* subunit interfaces. Experimental density is shown as a semi-transparent surface over the atomic model. Also see Figs. S2 and S3.

Mass photometry confirmed the structural predictions in the silkworm system (Fig. 5B). Mixing purified Spn-E with Siwi-piRNA, target RNA, GTSF1, and Mael resulted in a ~375 kDa peak corresponding to the Siwi*-Spn-E complex. The peak was lost upon omission of target RNA, GTSF1, or Mael, demonstrating that the fully intact Siwi* complex is necessary for recruiting Spn-E. This observation mirrors previous results showing that loss of Mael disrupts Siwi–Spn-E interactions and the production of Ago3-associated piRNAs in BmN4 cells⁴⁹.

Cryo-EM further confirmed the formation of the Siwi*-Spn-E complex (Fig. 5C). A reconstruction shows that the body of Spn-E binds adjacent to the Siwi MID domain, with its C-terminal ZnF domain extending into Siwi’s central RNA-binding cleft. This configuration positions Spn-E to capture the 3′ product of target cleavage, which is destined for loading into Ago3 as a new piRNA via the ping-pong cycle. Notably, despite its proposed helicase function⁵⁰, our Spn-E preparation did not exhibit helicase activity, nor did ATP influence the release of cleaved target RNA from Siwi. Thus, while GTSF1 and Mael are essential for recruiting Spn-E to Siwi, they do not activate its helicase function.

Finally, we examined the details of Spn-E recruitment. Spn-E contains an extended, 32-residue N-terminal region, which passes between Mael and Piwi and terminates as a short aliphatic α-helix bound to a hydrophobic patch at the Siwi-GTSF1 interface (Fig. 5D–E). These contacts explain the requirement of both GTSF1 and Mael for Spn-E recruitment. Indeed, deletion of the N-terminal region impaired recruitment to Siwi* (Fig. 5B). The combined results support the model that inter-subunit interfaces formed within Siwi* function as recruitment platforms for additional piRNA factors and assembly of larger silencing complexes.

Discussion

PIWI proteins and their associated piRNAs protect animal germ cells from transposons and other genomic threats by recognizing and silencing complementary RNAs. Despite decades of research, how piRNA-target engagement triggers a functional silencing response has remained unclear. Our results reveal the mechanism by which Siwi recruits cofactors GTSF1 and Mael upon sensing extended target pairing, defining the activated piRNA-directed silencing complex, Siwi*.

A central finding is that Siwi alone remains in a catalytically inactive state, even when bound to a fully complementary target RNA (Fig. 4A,B). Structural analyses show that sponge and mouse PIWI proteins also fail to adopt the active, “plugged-in”, conformation when engaging target RNAs alone^13,39. This contrasts sharply with bacterial Argonaute, which activates upon target binding^43,44,51, and eukaryotic Argonautes, which maintain a pre-formed active site^51–53. We propose that reluctance to activate is a conserved feature of PIWI proteins, contributing to their inherently slow catalytic activity^37,38.

Far from being a liability, this sluggishness may be an adaptive solution to a fundamental molecular challenge: how to recognize legitimate targets in a system where guide RNAs are long (25–35 nt) and mismatches are well-tolerated¹². A short stretch of base pairing is sufficient to stabilize PIWI-piRNA-target interactions^12,13, meaning that mere binding is not enough to distinguish functional targets. By requiring accessory proteins for activation, PIWI proteins enforce a higher standard: productive cleavage and downstream signaling occur only when pairing extends deeply into the piRNA 3’ region. GTSF1 and Mael directly contact these distal regions of the guide-target duplex, effectively reading out its full length. In this way, the PIWI* complex becomes a sensor that couples extended pairing to catalytic activation.

PIWI* assembly also appears to create a docking platform for downstream factors. One such factor is Spn-E, a putative RNA helicase essential for piRNA biogenesis. Our data show that Spn-E binds selectively to the fully assembled Siwi* complex (Fig. 5). The resulting configuration places Spn-E near the cleavage site, poised to receive the 3′ fragment of the cleaved target RNA. This positioning supports a model in which Spn-E captures the 3′ cleavage product for loading into a new PIWI protein, potentially initiating the ping-pong amplification cycle. While we did not detect helicase activity in recombinant Spn-E under our experimental conditions, its recruitment suggests a non-catalytic scaffolding role or a regulated helicase function that may require additional cofactors.

More broadly, AlphaFold-based predictions indicate the conservation of PIWI* assemblies across animals, from insects to mammals to basal metazoans. This finding, together with recent biochemical and genetic work in Drosophila by Brennecke and colleagues⁵⁴, supports the idea that PIWI* complexes are ancient and central components of the piRNA pathway. We suggest that PIWI* complexes act as molecular logic gates that integrate piRNA-target pairing information with the activation of PIWI catalysis and the recruitment of downstream silencing effectors, thereby linking recognition to response in a regulated manner.

Limitations of the Study

This study defines the structure and activation mechanism of the Siwi* complex using purified recombinant components. While our structural findings are supported by genetic and biochemical evidence from Drosophila, where Brennecke and colleagues demonstrate that Piwi forms a similar target-engaged complex⁵⁴, the conservation of PIWI* assembly and function across animal species remains to be fully established. Moreover, even where structural features are preserved, the requirements for PIWI* formation may vary between organisms or among PIWI paralogs. Finally, although our in vitro results align with observations from natural piRNA systems, the full physiological relevance of the Siwi* complex and its downstream recruitment functions remains to be validated in vivo.

Resource Availability

Lead contact:

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Ian J. MacRae (macrae@scripps.edu).

Materials Availability:

Expression plasmids for Siwi, GTSF1, Mael, and Spn-E are available upon request.

Data and Code Availability

• The cryo-EM maps and corresponding atomic model coordinates are deposited in the EMDB and Protein DATA Bank, respectively, as follows: SIWI-piRNA-target-GTSF1 (EMD-49407, 9NHB); SIWI-piRNA-target-GTSF1-Maelstrom (EMD-49408, 9NHC); SIWI-piRNA-target-GTSF1-Maelstrom, conformation-2 (EMD-49409, 9NHD); SIWI-piRNA-target-Maelstrom (EMD-49410, 9NHE); SIWI-piRNA-target-GTSF1-Maelstrom-Spindle-E (EMD-49422, 9NHS) and are publicly available as of the date of publication.

• All original code has been deposited at Figshare and is publicly available at DOI: 10.6084/m9.figshare.29815508 as of the date of publication.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

STAR Methods

METHOD DETAILS

Recombinant protein cloning, mutagenesis, and expression

DNA fragments encoding an N-terminal truncation of Siwi protein from Bombyx mori Siwi (NCIB: AB332313, residues 88–899) were cloned as an SfoI-XhoI fragment into a modified form of pFastBac HTA (Thermo Fisher) to generate expression plasmids for the Bac-to-Bac baculovirus expression system (Thermo Fisher) (Table S1). Baculoviruses were used to infect Sf9 cells at a density of 2×10⁶ cells per ml. Infected cells were incubated at 27°C, with gentle shaking to aerate, and harvested after 3 days.

The Bombyx mori GTSF1 coding sequence was cloned into an in-house bacterial expression plasmid. This plasmid contains an N-terminal maltose-binding protein (MBP) and 6xHis-tag, followed by the recognition site for tobacco etch virus (TEV) protease. The plasmid was transformed into BL21(DE3) cells (New England Biolabs) and transformed cells were grown in LB media supplemented with 1 μM ZnSO₄ at 37°C until OD₆₀₀ of 0.7–0.8. The protein expression was induced by adding IPTG to a final concentration of 1 mM and incubating shaking for 16 hours at 16 °C.

Bombyx mori

Mael was cloned into pET-21 vector containing a C-terminal TEV protease recognition site and 6xHis-tag. The Mael Δ1–88 deletion mutant was cloned with the NEBuilder HiFi DNA Assembly Master Mix (New England Biosciences) using PCR primers designed with the NEBuilder web tool. Plasmids were transformed into BL21(DE3) cells (New England Biolabs), and transformed cells were grown in LB at 37°C until OD₆₀₀ of 0.7–0.8. The protein expression was induced by adding IPTG to a final concentration of 1 mM and incubating shaking for 16 hours at 16 °C.

Bombyx mori

Spindle-E was cloned into pFastBac-Dual (Thermo Fisher Scientific) vector containing a C-terminal TEV protease recognition site and 6xHis-tag. Bacmid DNA was generated using the Bac-to-Bac system and used to transfect Sf9 cells to yield a baculovirus expressing the Spn-E. The Spn-EΔN (Δ2–52) construct was cloned with the NEBuilder HiFi DNA Assembly Master Mix (New England Biosciences) using PCR primers designed with the NEBuilder web tool. Baculoviruses were used to infect Sf9 cells at a density of 2×10⁶ cells per ml. Infected cells were incubated at 27°C, with gentle shaking to aerate, and harvested after 3 days.

Preparation of SIWI–guide RNA complexes

Bombyx mori

Siwi-guide complexes were purified as described previously¹³. Sf9 cell pellets were resuspended in lysis buffer (50 mM Tris pH 8, 300 mM NaCl, 0.5 mM TCEP) and lysed using a high-pressure (18,000 psi) microfluidizer (Microfluidics M110P). The lysate was clarified by centrifugation at 30000 × g for 20 min, followed by purification by Ni-NTA resin (Ni NTA agarose, Qiagen) for 1 hour at 4°C. Micrococcal nuclease (Takara) treatment was done to remove bound RNA in a buffer (50 mM Tris-HCl pH 8, 300 mM NaCl, 20 mM imidazole, 5 mM CaCl2, and 0.5 mM TCEP) for 45 minutes at room temperature. Following incubation, the resin was thoroughly washed three times with Ni-wash buffer (50 mM Tris, pH 8, 300 mM NaCl, 20 mM imidazole, 0.5 mM TCEP). The elution was done in three column volumes of Nickel Elution Buffer (50 mM Tris, pH 8, 300 mM NaCl, 300 mM imidazole, 0.5 mM TCEP). After elution, the protein was loaded with a synthetic guide RNA bearing a 5’-terminal phosphate and 2′-OCH3 group on its 3’ nucleotide (IDT). The His tag was removed from the protein using Tobacco Etch Virus (TEV) protease before it was dialyzed against Hi-Trap Dialysis Buffer (50 mM Tris, pH 8, 300 mM NaCl, 0.5 mM TCEP) overnight at 4°C. Siwi molecules loaded with the synthetic guide RNA were isolated using a biotinylated antisense oligonucleotide (IDT) immobilized on High-Capacity NeutrAvidin (Thermo Fisher). The bound resin was. Washed with 200 column volumes of Wash A (0.01% CHAPS, 2 mM magnesium acetate, 100 mM potassium acetate, 0.5mM TCEP, 30 mM Tris pH 8), 300 column volumes Wash B (0.01% CHAPS, 2 mM magnesium acetate, 2M potassium acetate, 0.5mM TCEP, 30 mM Tris pH 8), and 50 column volumes Wash C (0.01% CHAPS, 2 mM magnesium acetate, 1M potassium acetate, 0.5mM TCEP, 30 mM Tris pH 8) and eluted with a biotinylated DNA competitor (IDT). Excess competitor DNA was removed from the eluate by incubation with High Capacity NeutrAvidin for 20 minutes at room temperature. After another dialysis against dialysis Buffer (20 mM Tris, pH 8, 200 mM NaCl, 0.02% CHAPS, 0.5 mM TCEP), the protein was passed through Q Sepharose Fast Flow resin (Cytiva) to remove residual nucleic acids. Aliquots were concentrated using an Amicon Ultra-15 30K MWCO device (Millipore Sigma)and flash-frozen in liquid nitrogen and stored at −80 °C until their intended use.

Preparation of recombinant GTF1

The BL21(DE3) cells expressing GTSF1 were harvested by centrifugation, resuspended in lysis buffer (50 mM Tris pH 8, 300 mM NaCl, 0.5 mM TCEP), and passed through a high-pressure (18,000 psi) microfluidizer (Microfluidics M110P) to induce cell lysis. The lysate was clarified by centrifugation at 30,000× g for 20 minutes at 4°C and applied to Ni-NTA resin (Ni NTA agarose, Qiagen) for 1 hour. After incubation, the resin was washed three times with Ni-wash buffer (50 mM Tris, pH 8, 300 mM NaCl, 20 mM imidazole, 0.5 mM TCEP). The elution was done in three column volumes of Nickel Elution Buffer (50 mM Tris, pH 8, 300 mM NaCl, 300 mM imidazole, 0.5 mM TCEP). The N-terminal His6 and MBP tags were removed using TEV protease. For MBP-GTSF1, this step was omitted. The protein was dialyzed against Hi-Trap Dialysis Buffer (50 mM Tris, pH 8, 300 mM NaCl, 20 mM imidazole, 0.5 mM TCEP) overnight at 4 °C. An additional purification step was done by passing the dialyzed protein through a 5 ml Hi-Trap Chelating column (GE Healthcare). The unbound material was collected, concentrated, and further purified by size exclusion chromatography (SEC). For SEC, a Superdex 200 Increase 10/300 column (GE Healthcare) was used, which was equilibrated in a SEC buffer (50 mM Tris, pH 8, 300 mM NaCl, and 0.5 mM TCEP). The peak fractions were collected, and the purity was checked using SDS-PAGE. The purest fractions were collected and concentrated using an Amicon Ultra-15 10K MWCO device (Millipore Sigma) before being stored at −80°C.

Preparation of recombinant Maelstrom

Both Bombyx mori Mael and Mael-Δ1–88 were purified using a protocol nearly identical to that used for GTSF1. The only additional step involved is the treatment of the Mael-bound Ni NTA agarose resin with Micrococcal nuclease for 45 minutes at room temperature in a buffer (50 mM Tris-HCl pH 8, 300 mM NaCl, 20 mM imidazole, 5 mM CaCl2, and 0.5 mM TCEP) after the first Ni purification step, followed by washing steps.

Preparation of recombinant Spindle-E

For both Bombyx mori Spn-E and Spn-E-ΔN (Δ2–52), Sf9 cells were spun down and resuspended in lysis buffer (50 mM Tris, pH 8, 300 mM NaCl, 0.5 mM TCEP). The protein was purified with Ni-NTA resin in the same way as Siwi up until the dialysis step (and excluding the addition of guide RNA). After dialysis, the TEV protease-cleaved protein was passed through a HiTrap chelating column loaded with Ni²⁺ and eluted using wash buffer (50 mM Tris, pH 8, 300 mM NaCl, 0.5 mM TCEP) with a step gradient of 20 mM, 40 mM, 80 mM, 160 mM and 300 mM imidazole. Fractions containing TEV-cleaved Spn-E were pooled and diluted in wash buffer containing 100 mM NaCl. The protein was loaded on a HiTrap Q column and eluted using wash buffer with a linear gradient from 100 mM to 500 mM NaCl. The fractions containing the pure Spn-E were pooled and concentrated using an Amicon Ultra-15 100K MWCO device (Millipore Sigma). Filtration buffer (50 mM Tris, pH 8, 300 mM NaCl, 0.5 mM TCEP) was added to the filtrate and centrifuged again in the concentrator. This step was repeated two more times before aliquoting, flash freezing in liquid nitrogen, and then storing at −80°C until use.

Mass Photometry

Purified Siwi, MBP-GTSF1, and Mael were mixed at concentrations of 10 nM Siwi, 40 nM GTSF1 and 40 nM Mael in presence of 30 nM target RNA (Lock-in Target 39) at room temperature for 15 minutes in Binding Buffer (50 mM Tris pH 8.0, 100 mM NaCl, 0.5 mM TCEP). Samples were diluted by half before being applied to the mass photometer for a final tested concentration of 5 nM Siwi, 20 nM GTSF1, 20 nM Mael and 15 nM target RNA. Complexes involving Siwi, MBP-GTSF1, Mael, and Spn-E were formed at 10 nM Siwi, 30 nM MBP-GTSF1, 30 nM Mael, and 30 nM Spn-E in the presence of 30 nM target RNA (Lock-in Target 39) at room temperature for 15 minutes in the same binding buffer. Samples were diluted by half before being applied for detection for a final tested concentration of 5 nM Siwi, 15 nM MBP-GTSF1, 15 nM Mael, 15 nM Spn-E and 15 nM target RNA. The size distributions of formed complexes were assessed using the Refeyn Two MP mass photometer. Data were analyzed using Refeyn DiscoverMP.

Target Cleavage Assay

Purified Siwi–guide RNA complexes (10 nM, final concentration) were incubated at 27 °C with complementary 5′-³²P-labelled target RNAs (1 nM, final concentrations) in a reaction buffer composed of 20 mM Tris pH 8.0, 100 mM NaCl, 2 mM MgCl₂, 0.5 mM TCEP, and 0.01 mg/ml baker’s yeast tRNA. The cleavage reaction was terminated at various intervals by mixing aliquots of each reaction with an equal volume of denaturing gel loading buffer (98% w/v formamide, 0.025% xylene cyanol, 0.025% w/v bromophenol blue, and 10 mM EDTA pH 8.0). 15% denaturing urea PAGE was used to resolve intact and cleaved target RNAs, which were subsequently visualized by phosphorimaging. The signals were quantified using ImageQuant TL (GE Healthcare).

Target RNA Cleave-n’-Seq library design and in vitro transcription

DNA oligos encoding target sites were designed with a constant region spanning positions 1–15 of the piRNA guide, followed by 14 randomized positions of either A/T, or C/G. Based on the length of the guide piRNA used (25 nt), only 10 of these 14 positions were engaged by the piRNA. The target sequence was flanked by two stretches of 5 A, followed by adapter sequences employed by the Scripps Genomics core, and finally, an upstream T7 promoter. The DNA was obtained from IDT, and the single-stranded template was amplified by PCR. The PCR product was used for T7 in vitro transcription, followed by purification using the Monarch RNA cleanup kit (NEB). The purified RNA was diluted to 100 nM, and two DNA blocking sequences spanning the adapter region were annealed at 1.2x molar concentration in annealing buffer (50 mM Tris-HCl pH 8, 100 mM NaCl) by heating 1 min at 90 °C, followed by 5 min at 22 °C, and finally storage on ice. This stock target RNA library was diluted to 10 nM final concentration, aliquoted, and stored at −80 °C.

Cleave-n’-Seq reactions and library preparation

Cleavage reactions were performed in the same reaction buffer as above, in 50 μl final volume with 1 nM Target, 10 nM Siwi, and a range of conditions of GTSF1 and Maelstrom (GTSF1 at 1 nM, 10 nM, 100 nM, or 1000 nM; Maelstrom at 250 nM; GTSF1 at 10 nM and Mael at 250 nM). A Siwi “no cleavage” reaction was also performed by omitting MgCl₂ from the reaction mixture. Reactions were stopped by the addition of 50 μl stop buffer (50 mM Tris-HCl pH 8, 100 mM NaCl, 25 mM EDTA, 0.05% SDS, 0.8 U Proteinase K) and purified using the RNA Clean & Concentrator kit (Zymo Research). The reverse transcription (RT) was performed with 1 μl of purified RNA using Superscript VI (Thermo Fisher) in 10 μl total reaction volume. The RT primer was added to the RNA samples, followed by annealing (1 min at 90 °C, 5 min at 65 °C, 1 min on ice), before adding the remaining components. The RT reaction was performed for 10 min at 55 °C, followed by 10 min at 80 °C. The resulting cDNA was amplified by PCR with KAPA HiFi Hotstart Ready mix (Roche), with an initial denaturation at 95 °C for 3 minutes, followed by 12 amplification cycles ( 20 s at 98 °C, 15 s at 66 °C, 15 s at 72 °C), and a final extension of 1 min at 72 °C. The resulting libraries were submitted to the Genomics core for sequencing on an AVITI sequencer (Element Biosciences).

Cleave-n’-Seq data analysis

Adapter sequences were removed with Trimmomatic⁵⁵ and reads for each sequence in the library were determined using a custom python script, without allowing mismatches, insertions, or deletions. Normalized reads were computed for each dataset, and the Siwi no cleavage reaction was used for normalization of the abundance of individual sequences in the library input. For each dataset, the peak of uncleaved sequences was determined with a custom R script, and used for scaling, and cleaved sequences were defined as having a relative abundance of −3 sigma from the peak or lower. The same script was used to characterize the effect of GSTF1 (Siwi-GTSF1vs. Siwi) and Mael (Siwi-GTSF1-Mael vs. Siwi-GTSF1). The peak identified the region of sequences with comparable cleavage between the two conditions, −3 sigma, and −6 sigma cutoffs from the peak were used to determine bins of sequences. For each sequence bin, the frequency of a match between the target and the piRNA guide at every position was computed by a custom R script. Dot plots for the data were generated in Prism (Graphpad). Statistical analysis of the sequence bin comparisons was performed by one way ANOVA, followed by Bonferroni’s multiple comparisons test. P values are indicated in the plots.

Grid Preparation for Cryo-EM

The complexes of Siwi-piRNA with target RNA and other proteins for the Cryo-EM study were prepared as follows. The Siwi-piRNA-target-GTSF1, and Siwi-piRNA-target-GTSF1-Maelstrom (Siwi*) complexes were formed at room temperature for 10 minutes in storage buffer (50 mM TRIS, pH 8.0, 100 mM NaCl, and 0.5 mM TCEP) at 1:1.5:2 and 1:2:5, molar ratio, respectively. The Siwi*-conformation 2, Siwi-piRNA-target-Maelstrom, and Siwi-piRNA-target-GTSF1-Maelstrom-Spindle-E (Siwi*-Spn-E) complexes were formed, using target RNA with Phosphorothioate modification at target position eleven, at room temperature for 10 minutes in buffer (30 mM Tris, pH 8.0, 100 mM KoAc, 2 mM Mg(OAc)₂, 0.5 mM TCEP) at 1:2:5, 1:2:5, and 1:2:5:2.5:2 molar ratio, respectively. Siwi-piRNA-target-GTSF1, Siwi*, Siwi*-conformation 2, Siwi-piRNA-target-Maelstrom, and Siwi*-Spn-E complexes were prepared using target 25-nt, 49-nt, 49-nt, 49-nt, and 39-nt respectively, yielding final complex concentrations of 0.89 mg/mL, 0.86 mg/mL, 0.86 mg/mL, 1.13 mg/mL, and 1.11 mg/mL, respectively.

For grid preparation, 300 mesh R1.2/1.3 UltraAuFoil Holey Gold grids (Quantifoil) were subjected to glow discharge under vacuum conditions for 30 seconds at a current of 15 mA using a Pelco easiGlow 91000 Glow Discharge Cleaning System (Ted Pella Inc.). Subsequently, 3 μL of the sample was applied to each grid surface, followed by blotting with Whatman #1 filter paper (55 mm diameter) for 4 seconds with a blot force of 4 on a Vitrobot Mark IV (Thermo Fisher Scientific) operating at 100% humidity and 4°C. The grids were then plunge-frozen in liquid ethane.

Cryo-EM Data Acquisition

The cryo-EM data of the complexes were collected on a TALOS Arctica transmission electron microscope (Thermo Fisher Scientific) with a field emission gun operating at an accelerating voltage of 200 kV and equipped with a K2 Summit (Gatan) direct electron detector, operated in electron-counting mode, using the automated data collection software Leginon⁵⁶ by image shift-based movements from the center of four adjacent holes to target the center of each hole for exposures. Each micrograph for the Siwi-piRNA-target-GTSF1, Siwi*, Siwi*-conformation 2, Siwi-piRNA-target-Maelstrom, and Siwi*-Spn-E complexes were collected as 23, 22, 26, 26, and 23 dose-fractionated movie frames, respectively over 4.6 s, 4.4 s, 5.2 s, 5.2 s, and 4.5 s, respectively and with a cumulative electron exposure of 50 e⁻ Å⁻².

The datasets were collected with a C2 aperture of 50 mm, at a magnification of 45000x, corresponding to 0.91 Å per pixel on the detector with a defocus range of −1.0 to −3.0 mm. Appion⁵⁷ image processing was used to run or micrograph frame alignment and dose-weighting in real time during data collection. All subsequent processings were performed in CryoSPARC⁵⁸. A total of 4687, 5281, 4802, 4991, and 3412 micrographs were collected with the Siwi-piRNA-target-GTSF1, Siwi*, Siwi*-conformation 2, Siwi-piRNA-target-Maelstrom, and Siwi*-Spn-E complexes, respectively.

Image Processing and 3D Reconstruction

For Siwi-piRNA-target-GTSF1 complex data processing, motion-corrected micrographs were subjected to patch CTF estimation, followed by blob picking, and two rounds of 2D classifications. Based on the similarity to the known Siwi structure and excluding smaller classes of junk particles, selected 2D class averages were subjected to ab initio reconstruction to generate four unique classes. The best-looking Siwi class was used to create templates for template picking. A total of 2,411,587 particles were extracted with binning 2×2 (1.48 Å per pixel, 130 pixels box size) and subjected to two rounds of 2D classifications. Based on the similarity to the known Siwi structure and excluding smaller classes of junk particles, 1,505,526 particles were selected from the 2D class averages and were subjected to ab initio reconstruction to generate six unique classes. These classes were used for 3D heterogeneous refinement. The best-looking Siwi class/classes, chosen from every round, were subjected to the next round of heterogeneous refinement in an iterative process. A total of three rounds of heterogeneous refinements were performed, and the particles corresponding to 3D maps resembling Siwi with piRNA-target RNA duplex were combined and subjected to homogeneous refinement, followed by two rounds of reference-free 3D classifications. The final Siwi class with a complete guide-target RNA duplex contained 27% particles. Homogeneous refinement and local refinement were performed on the final particle stack, which was re-extracted with no binning (0.91 Å per pixel, 300 pixels box size) from micrographs with an estimated CTF fit to a resolution of 5 Å or better. The final 161,717 particles were subjected to final local refinement. This resulted in a final map of the Siwi-piRNA-target-GTSF1 complex with a resolution of 3.7 Å, determined by gold-standard FSC at a cutoff of 0.143. Local resolution estimation was performed, and the estimated resolution range was 3.1–7.1 Å. The GTSF1 density was not observed in this complex except for GTSF1 residues 101–104.

For Siwi* complex data processing, motion-corrected micrographs were subjected to patch CTF estimation, followed by blob picking. A total of 1,470,893 particles were extracted with binning 2×2 (1.48 Å per pixel, 150 pixels box size) and subjected to two rounds of 2D classifications. Based on resolved 2D classes and excluding smaller classes of junk particles, 1,105,389 particles were selected from 2D class averages. From the selected particle stack, 52,000 particles were subjected to ab initio reconstruction to generate three unique classes. These classes were used for 3D heterogeneous refinement. The best-looking Siwi* class, chosen from every round, was subjected to the next round of heterogeneous refinement along with a new stack of particles from the collected data in an iterative process for three rounds until the classes with junk particles contained 4–5% particles of the total particle stack of corresponding heterogeneous refinement. At the end of the last round of heterogeneous refinement, the 3D map resembling the Siwi* class was subjected to homogeneous refinement, followed by a reference-free 3D classification. The final best-looking Siwi* class from 3D classification contained 43% of the particles. Homogeneous refinement was performed on the final particle stack, which was re-extracted with no binning (0.91 Å per pixel, 300 pixels box size) from micrographs with an estimated CTF fit to a resolution of 5 Å or better. The final 198,296 particles were subjected to homogeneous and local refinement, followed by CTF refinements and final local refinement with a mask of the whole complex. This resulted in a final map of the Siwi* complex with a resolution of 3.6 Å, determined by gold-standard FSC at a cutoff of 0.143. Local resolution estimation was performed, and the estimated resolution range was 3.0–5.9 Å.

For Siwi*-conformation 2 complex data processing, motion-corrected micrographs were subjected to patch CTF estimation, followed by template picking. Templates for picking were chosen from 2D classes from the Siwi* workflow discussed above. A total of 2,346,312 particles were extracted with binning 2×2 (1.48 Å per pixel, 150 pixels box size) and subjected to three rounds of 2D classifications. Based on the resolved 2D classes and excluding smaller classes of junk particles, 795,224 particles were selected from the 2D class averages and were subjected to ab initio reconstruction to generate three unique classes. These classes were used for 3D heterogeneous refinement. The best-looking Siwi* class, chosen from every round, was subjected to the next round of heterogeneous refinement in an iterative process. A total of three rounds of heterogeneous refinements were performed, and the particles corresponding to 3D maps resembling the Siwi* class were subjected to non-uniform refinement, followed by local refinement. A 3D variability was performed on a mask around the GTSF1 N-terminal domain. In 3D variability analysis cluster mode, four classes were generated and subjected to non-uniform refinement, followed by local refinement. Three among four classes were reconstructed with a complete map of GTSF1 (conformation 1) whereas, the fourth class was reconstructed with missing density for the GTSF1 N-terminal domain (Siwi*-conformation 2). The particle stacks of conformation 1, combined containing 215,217 particles, were subjected to final local refinement with a mask of the whole complex and resulted in a final map of the Siwi* complex with a resolution of 3.8 Å (conformation 1, the same map was obtained in the Siwi* complex data processing with a resolution of 3.6 Å discussed above). In this workflow, Siwi*-conformation 2 was analyzed as the final map. The final particle stack of Siwi*-conformation 2, which was re-extracted with no binning (0.91 Å per pixel, 300 pixels box size) from micrographs with an estimated CTF fit to a resolution of 15 Å or better. The final 90,700 particles were subjected to final local refinement with a mask of the whole complex. This resulted in a final map of the Siwi*-conformation 2 complex with a resolution of 4.1 Å, determined by gold-standard FSC at a cutoff of 0.143. Local resolution estimation was performed, and the estimated resolution range was 3.4–8.9 Å. A Siwi-piRNA-target-GTSF1 (The GTSF1 density was not observed in this complex except for GTSF1 residues 101–104) class was also obtained as one of the ab initio classes. This Siwi-piRNA-target-GTSF1 class, chosen from every round, was subjected to the next round of heterogeneous refinement in an iterative process for three rounds. The final particle stack of Siwi-piRNA-target-GTSF1 class contained 139,866 particles, subjected to non-uniform refinement followed by a final local refinement, were resulted in a final map of Siwi-piRNA-target-GTSF1 complex with a resolution of 4.0 Å (same map was obtained with a resolution of 3.7 Å in Siwi-piRNA-target-GTSF1 complex data processing discussed above).

For Siwi-piRNA-target-Maelstrom complex data processing, motion-corrected micrographs were subjected to patch CTF estimation, followed by blob picking (for the first 279 images), and two rounds of 2D classifications. Based on resolved 2D classes and excluding smaller classes of junk particles, selected 2D class averages were subjected as templates for template picking. A total of 2,268,300 particles were extracted with binning 2×2 (1.48 Å per pixel, 150 pixels box size) and subjected to two rounds of 2D classifications. Based on the resolved 2D classes and excluding smaller classes of junk particles, 397,459 particles were selected from the 2D class averages and were subjected to ab initio reconstruction to generate three unique classes. These classes were used for 3D heterogeneous refinement. The best-looking Siwi-piRNA-target-Maelstrom class, chosen from every round, was subjected to the next round of heterogeneous refinement in an iterative process for two rounds until the classes with junk particles contained 11–7% particles of the total particle stack of corresponding heterogeneous refinement. At the end of the last round of heterogeneous refinement, the final best-looking Siwi-piRNA-target-Maelstrom class contained 82% of the particles. Homogeneous refinement followed by local refinement was performed on the final particle stack, which was re-extracted with no binning (0.91 Å per pixel, 300 pixels box size) from micrographs with an estimated CTF fit to a resolution of 5 Å or better. The final 158,398 particles were subjected to final local refinement. This resulted in a final map of the Siwi* complex with a resolution of 4.2 Å, determined by gold-standard FSC at a cutoff of 0.143. Local resolution estimation was performed, and the estimated resolution range was 3.4–7.7 Å.

For Siwi*-Spn-E complex data processing, motion-corrected micrographs were subjected to patch CTF estimation, followed by blob picking. A total of 611,734 particles were extracted with binning 2×2 (1.48 Å per pixel, 208 pixels box size) and subjected to two rounds of 2D classifications. Based on resolved 2D classes and excluding smaller classes of junk particles, 530,244 particles were selected from 2D class averages. From the selected particle stack, 70,849 particles were subjected to ab initio reconstruction to generate four unique classes. These classes were used for 3D heterogeneous refinement. The best-looking Siwi*-Spn-E class, chosen from every round, was subjected to the next round of heterogeneous refinement along with a new stack of particles from the collected data in an iterative process for two rounds until the classes with junk particles contained 18–15% particles of the total particle stack of corresponding heterogeneous refinement. At the end of the last round of heterogeneous refinement, the best-looking Siwi*-Spn-E class contained 51% of the particles and was subjected to non-uniform refinement, followed by a local refinement. This final stack of 83,871 particles was further subjected to local refinement with a mask around the Siwi* region of the complex, followed by final local refinement with a mask of the whole complex. This resulted in a final map of Siwi*-Spn-E complex with a resolution of 4.2 Å, determined by gold-standard FSC at a cutoff of 0.143. Local resolution estimation was performed, and the estimated resolution range was 4.0–9.0 Å. The sphericity and global resolution of the final maps were evaluated using the Remote 3DFSC Processing Server (https://3dfsc.salk.edu)⁵⁹.

Further image processing and 3D reconstruction details are available in Methods S1.

Model Building and Refinement

For Siwi-piRNA-target-GTSF1, Siwi*, Siwi*-conformation 2, Siwi-piRNA-target-Maelstrom, and Siwi*-Spn-E complex model building, the initial model of Siwi only was obtained from the crystal structure of Siwi bound to piRNA (PDB ID: 5GUH). The piRNA-target RNA duplex was modeled in Coot model building tool⁶⁰. The initial model for GTSF1, Maelstrom, and Spindle-E was obtained using AlphaFold2²⁷. The domains of proteins and the piRNA-target RNA duplex were first fitted in the final cryo-EM map as a rigid body in ChimeraX^61,62, followed by a rigid body refinement in Phenix⁶³ to generate the initial model. The real-space refinement of the initial model was performed optimizing global minimization, atomic displacement parameters, and local grid search using PHENIX⁶³ Iterative rounds of manual building and fixing of geometric, rotamer, and Ramachandran outliers were performed in Coot and ISOLDE⁶⁴ in ChimeraX. The final model was validated by comprehensive validation (cryo-EM) and EMRinger⁶⁵ suite in Phenix.

QUANTIFICATION AND STATISTICAL ANALYSIS

Calculation of Interchain PAE (iPAE) Scores

For each protein complex, an iPAE-like score was calculated using the predicted alignment error (PAE) matrices generated by ColabFold (Table S2)²⁸. The scoring system was heuristically designed to elevate complexes with a large number of high-confidence PAEs and punish those lacking high-confidence PAEs.

For each complex, residue indices were delineated using chain-specific boundaries determined from the chain lengths. For each model, the interchain PAEs evaluated involved the final chain with all the other chains of the complex. For example, for complexes with two chains (A and B), interchain PAEs between B and A were considered. For complexes with three chains (A, B, and C), interchain PAEs between C and A, and C and B were used, allowing assessment of the confidence of C interactions with A and B.

For each model, the Colabfold PAE matrix provided a predicted error value $d$ for every cross-chain residue pair. Each pair was assigned a weighted score defined as:

$$s=P(d)\times w(d)$$

where the probability weight $P(d)$ is given by:

$$P(d)=\left\{\begin{array}{l}4.0,\text{if}d\le 2\\ 2.0,\text{if}d\le 5\\ 0.0\text{if}d>5\end{array}\right.$$

and the TM-score-like weight $w(d)$ is defined as:

$$w(d)=\frac{1}{1+{\left(\frac{d}{d_0}\right)}^2}$$

with $d_0$ set to 8.0 by default. Thus, residue pairs with high PAE values $(d>5)$ were effectively excluded from contributing to the interchain score.

For each residue pair, weighted scores from the two highest-ranked models were aggregated. The aggregation involved computing the mean of the nonzero scores and scaling it by a consistency factor, defined as the fraction of models in which the residue pair received a nonzero score. For example, if a residue pair was scored in only one of the two models, the consistency factor was 0.5.

A raw iPAE score was obtained by summing the averaged scores over all pairs and normalizing by $N+{d_0}^{\prime }$, where $N$ denotes the total number of residue pairs with nonzero aggregated scores and ${d_0}^{\prime }$ is a length-corrected normalization factor:

$${d^{\prime }}_0=1.24N^{1/3}-1.8$$

Hence, the raw score was computed as:

$$\mathit{raw}\mathit{score}=\frac{\sum _{i=1}^N{s}_i}{N+{d^{\prime }}_0}$$

To account for the possibility of undersized interfaces, an interface scaling factor was applied. This factor was defined as:

$$\mathit{scaling}\mathit{factor}=\mathrm{m}\mathrm{i}\mathrm{n}\left(1,{\left(\frac{N}{\mathrm{10,000}}\right)}^{0.3}\right)$$

thereby penalizing interfaces with a low number of residue pairs.

The final iPAE score was obtained by multiplying the raw score by the interface scaling factor:

$$iPAEscore=rawscore\times scalingfactor$$

This procedure integrates both the magnitude and the consistency of the PAE values across models, while also incorporating corrections for interface size and chain length. The resulting iPAE score provides a quantitative measure of the confidence in the predicted interchain interactions. The score has no upper limit, but we empirically found that moderate-confidence scores are in the range of 0.4–1.1, and high-confidence scores have a score above 1.2.

Normalization of Cleave-’n-Seq data

Reads from each Cleave-’n-Seq experiment were first normalized to an uncleaved control library to account for abundance differences intrinsic to the input library. A normal distribution was fit over the uncleaved sequences of each experiment. This provided a quantitative measure of the noise in the data and, by scaling each data set to its calculated peak, an unbiased way of placing all data sets on the same scale. Uncleaved sequences were defined as read ≥3 standard deviations below the mean of each normalized fit (see Methods S1).

Supplementary Material

3

Table S2. Colabfold inputs, related to STAR Methods

Tab 1: Organism names and Uniprot IDs for the core piRNA factors screened. PIWI sequences used for each organism are indicated in the colored table on the right. Tab 2: Uniprot IDs and FlyBase designations of Drosophila piRNA factor sequences screened against Piwi, Aub, and Ago3.

Key resources table.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Bacterial and virus strains
E. coli BL21(DE3)	New England Biolabs	Cat# C2527I; RRID: N/A
E. coli OmniMAX	Thermo Fisher	Cat# C854003; RRID: N/A
Baculovirus: His6-TEV-Siwi	This study	N/A
Baculovirus: Spn-E-His6	This study	N/A
Biological samples
Chemicals, peptides, and recombinant proteins
NEBuilder HiFi DNA Assembly Master Mix	New England Biolabs	E2621S
TEV Protease	In house	N/A
Micrococcal nuclease	Takara	2910A
KAPA HiFi Hotstart Ready mix	Roche	07958927001
Superscript VI	Thermo Fisher	18090010
Proteinase K	New England Biolabs	P8107S
Critical commercial assays
Deposited data
SIWI-piRNA-target-GTSF1	This study	EMD-49407, 9NHB
SIWI-piRNA-target-GTSF1-Maelstrom	This study	EMD-49408, 9NHC
SIWI-piRNA-target-GTSF1-Maelstrom, conformation-2	This study	EMD-49409, 9NHD
SIWI-piRNA-target-Maelstrom	This study	EMD-49410, 9NHE
SIWI-piRNA-target-GTSF1-Maelstrom-Spindle-E	This study	EMD-49422, 9NHS
Crystal structure of silkworm PIWI-clade Argonaute Siwi bound to piRNA	Matsumoto, et al.40	5GUH
Cryo-EM Structure of MILI-piRNA-target (26-nt)	Li, et al.39	9IJ3
Cryo-EM Structure of EfPiwi-piRNA-target (25-nt, bilobed)	Li, et al.39	9IIY
Siwi, Siwi-GTSF1, Siwi* RNA Bind-n’-Seq data	This study	PRJNA1301092
Experimental models: Cell lines
Sf9 insect cells (Spodoptera frugiperda ovarian cells)	Expression Systems	Cat# 94–001S; RRID:CVCL_0549
Experimental models: Organisms/strains
Oligonucleotides
Oligonucleotide sequences, see Table S1	This study	N/A
Recombinant DNA
pFastBac His6-Flag-TEV Siwi	This study	N/A
pFastbac-Dual Spn-E-His6	Gene Universal (this study)	N/A
pet21(+) His6-TEV-Mael	Twist Bioscience (this study)	N/A
pSV272 GTSF1	This study	N/A
pFastBac HTA	Thermo Fisher	Cat# 10584027
Software and algorithms
DiscoverMP	Refeyn	N/A
Trimmomatic	Bolger, et al.55	https://github.com/timflutre/trimmomatic
Prism	Graphpad	https://www.graphpad.com/features
Appion	Lander, et al.57	https://github.com/leginon-org/leginon
CryoSPARC	Punjani, et al.59	https://cryosparc.com/
MotionCor2	Zheng, et al.58	https://emcore.ucsf.edu/ucsf-software
3DFSC	Tan, et al.60	https://3dfsc.salk.edu
Coot	Emsley, et al.61	https://www2.mrc-lmb.cam.ac.uk/personal/pemsley/coot/
Phenix	Liebschner, et al.64	https://phenix-online.org/
ChimeraX	Pettersen, et al.63	https://www.cgl.ucsf.edu/chimerax/
AlphaFold2	Jumper, et al.27	https://github.com/google-deepmind/alphafold
ISOLDE	Croll, et al.65	https://tristanic.github.io/isolde/
EMRinger	Barad, et al.66	https://github.com/fraser-lab/EMRinger
ColabFold	Mirdita, et al.28	https://github.com/sokrypton/ColabFold
calc_iPAE.py	This study.	DOI: 10.6084/m9.figshare.29815508
Other

Highlights.

• Extended piRNA-target pairing triggers assembly of the Siwi* silencing complex

• Siwi* forms via GTSF1 and Maelstrom recruitment

• Siwi* rapidly cleaves target RNAs and recruits downstream effectors

• A conserved PIWI* architecture underlies transposon silencing across animals