MIWI N-terminal arginines orchestrate generation of functional pachytene piRNAs and spermiogenesis

Abstract N-terminal arginine (NTR) methylation is a conserved feature of PIWI proteins, which are central components of the PIWI-interacting RNA (piRNA) pathway. The significance and precise function of PIWI NTR methylation in mammals remains unknown. In mice, PIWI NTRs bind Tudor domain containing proteins (TDRDs) that have essential roles in piRNA biogenesis and the formation of the chromatoid body. Using mouse MIWI (PIWIL1) as paradigm, we demonstrate that the NTRs are essential for spermatogenesis through the regulation of transposons and gene expression. The loss of TDRD5 and TDRKH interaction with MIWI results in attenuation of piRNA amplification. We find that piRNA amplification is necessary for transposon control and for sustaining piRNA levels including select, nonconserved, pachytene piRNAs that target specific mRNAs required for spermatogenesis. Our findings support the notion that the vast majority of pachytene piRNAs are dispensable, acting as self-serving genetic elements that rely for propagation on MIWI piRNA amplification. MIWI-NTRs also mediate interactions with TDRD6 that are necessary for chromatoid body compaction. Furthermore, MIWI-NTRs promote stabilization of spermiogenic transcripts that drive nuclear compaction, which is essential for sperm formation. In summary, the NTRs underpin the diversification of MIWI protein function.

The N-terminus of PIWI proteins contains arginines (NTRs) that are symmetrically dimethylated (sDMA) by protein arginine methyltrasferase 5 (PRMT5) ( 29 ) and bind to Tudor domain containing proteins (TDRDs) ( 21 , 30 , 22 , 31 , 32 ).Multiple TDRDs interact with mouse PIWI proteins and have essential roles in piRNA biogenesis and function ( 33 , 20 , 34 , 23 , 35-39 ).TDRD1 interacts with MILI and is involved in selecting appropriate piRNA precursors for MILI piRNAs and MIWI2 loading ( 34 ).TDRKH (TDRD2) contains a transmembrane domain that tethers it to the cytoplasmic surface of mitochondria, and is essential for 3 end trimming of MILI-bound prepachytene piRNAs ( 35 ) and for production of all MIWI-bound piRNAs ( 38 ).TDRKH interacts directly with MIWI, preferentially recognizing its unmethylated arginines ( 39 ).TDRD5 binds MIWI and piRNA precursors and is required for piRNA production from the entire length of the precursor transcripts ( 37 ).The sDMAs of MIWI are essential for mediating direct interactions to Tudor domains of TDRD6 ( 22 ).Deletion of TDRD6 leads to arrest at the elongation phase of haploid spermatids (step 12) ( 20 ,23 ).TDRD6 does not have critical functions in piRNA biogenesis or piRNA-mediated transposon control but rather has a direct role in other CB-related functions of MIWI ( 23 ).How MIWI can coordinate such diverse functions remains unknown.
Here, we report that MIWI-NTRs, by coordinating interactions with TDRDs required for piRNA biogenesis, orchestrate the functions of MIWI and piRNAs.Surprisingly, we find that the loss of TDRD5 and TDRKH interaction with MIWI results in defective piRNA amplification, rather than an expected failure of piRNA biogenesis.We demonstrate that MIWI piRNA amplification is necessary for transposon control and for sustaining piRNA levels including select, pachytene piRNAs that target specific mRNAs required for spermiogenesis.Our findings suggest that the vast majority of pachytene piRNAs are dispensable, acting as self-serving genetic elements that rely on MIWI-mediated piRNA amplification for their propagation.MIWI-NTRs also mediate interactions with TDRD6 that are necessary for CB compaction.Furthermore, MIWI-NTRs underlie stabilization of spermiogenic transcripts that drive nuclear compaction, which is essential for sperm formation.In summary, MIWI-NTRs enable functional diversification through the interaction with distinct TDRD proteins.

Generation and genotyping of the Miwi RK allele
The mouse Miwi RK allele was generated through CRISPR-CAS9 gene editing on fertilized 1-cell stage zygotes of B6CBAF1 / Crl genetic background as previously described ( 40 ,41 ).Cas9 mRNA was injected together with a small guide RNA (CTGACCTCGTGCCCTGCCGC) targeting the MIWI-NTR encoding exon plus a 200 nt ssDNA donor oligonucleotide (GTGCTTT AAAAAGGTTT AGTGAT A A AATGGTGAATGC ACGTGAGCCC ATCGTGTGCTT TTCCTCTCTGA CA GAAAATGA CTGGCaagGCCaagG CTaaaGCCaagGGCaaaGC AaagGGTC AGGA GA CGGTG CAGCA TGTTGGGGCTGCTGCGGTGAGT ACCA TTCTT A TA TAGCTCAA TAGC ATCTTAAC AACC AGCC A) delivering the expected nucleotide changes flanked by two 5 and 3 homology arms, 84 nt and 83 nt in length, respectively ( Supplementary Figure S1 A, B).F 0 offspring carrying the Miwi RK allele were identified by PCR and validated with Sanger sequencing ( Supplementary Figure S1 C, D).Miwi RK line was established from one F 0 founder mouse and was back-crossed several times onto a C57BL6 / N genetic background.Therefore, transgenic mice in this study, were maintained on a mixed B6CBAF1 / Crl; C57BL6 / N genetic background.

Mouse experimentation
Male fertility was assayed by timed matings of an adult stud male with a wild-type C57BL6 / N adult female for 4 days and separation of the female upon detection of a vaginal plug.The plug date was counted as embryonic day 0.5 (E0.5) and the number of embryos at E16.5 was counted for each plugged female.Animal tissue samples were collected from one or more litters per experiment and allocated to groups according to genotype.No further randomization or blinding was applied during data acquisition and analysis.Animals were maintained at the University of Edinburgh, UK, in accordance with the regulation of the UK Home Office.Ethical approval for the mouse experimentation has been given by the University of Edinburgh's Animal Welfare and Ethical Review Body and the work was done under license from the United Kingdom's Home Office.Furthermore, mouse experimentation for this project was reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of the University of Pennsylvania.

Histology and immunohistochemistry
For periodic acid Schiff (PAS) staining, testes and epididymis were fixed overnight in Bouin's fixative (Millipore Sigma) followed by several washes in 70% ethanol and embedding in paraffin.Next, 3 μm sections were cut on a microtome (Leica), rehydrated through an alcohol series according to standard laboratory procedures and then stained with a PAS staining kit (TCS Biosciences) according to the manufacturer's instructions.Stained sections were rehydrated through a reverse alcohol series and mounted on coverslips with Pertex mounting medium (Pioneer Research Chemicals) according to standard laboratory procedures.Slides were imaged on Zeiss AxioScan scanning microscope at 40 × magnification.Cropped images were exported using the Zeiss Zen software and further processed in ImageJ.
For IF, testes were fixed overnight in 4% PFA and embedded in paraffin.The testicular tissue was cut into 5 μm thick sections on a microtome (Leica), mounted onto Superfrost Plus microscope slides (Fisherbrand) and left to dry overnight at 47 • C. The tissue was dewaxed by immersion in xylene three times for 5 min each, rehydrated sequentially in 100% ethanol and finally rinsed in distilled water (dH 2 O).Slides were placed in Coplin jars filled with antigen retrieval solution (10 mM sodium citrate buffer, pH 6.0) and boiled in microwave oven for 10 min.After cooling down for 30 min, sections were washed twice in dH 2 O.A hydrophobic barrier pen was used to circle the tissue section on the slide before blocking with 5% normal goat serum (NGS) in PBS, for 1 h at room temperature.Subsequently, sections were incubated overnight with primary antibodies diluted in 5% NGS and 0.3% Triton-X in PBS, inside humidified chamber at 4 • C. Next day, the sections were washed in PBS three times, with gentle agitation, for 10 min each.Secondary antibody incubation took place inside lightprotected jar containing 5% NGS and 1 μg / ml DAPI in PBS, for 1 h at room temperature.Lastly, sections were washed in PBS, three times for 15 min each, covered with fluorescencepreserving medium and sealed with nail polish.The antibodies used for IF are listed in Supplementary Table S13 .IF images were acquired with a Leica TCS confocal microscope at 63 × magnification.

Electron microscopy
P24 testes were isolated and fixed in 100 mM sodium cacodylate pH 7.4 with 2% paraformaldehyde and 2.5% glutaraldehyde.Samples were processed and imaged on JEOL JEM-1010 transmission electron microscope, at the Electron Microscopy Resource Laboratory at the University of Pennsylvania.

Lysate preparation, immunoprecipitation and western blot
P24 testes were homogenized with a pestle in ice cold lysis buffer (20 mM Tris pH 7.5, 200 mM NaCl, 2.5 mM MgCl 2 , 0.5% NP-40, 0.1% Triton X-100 and 1 mM TCEP) supplemented with EDTA-free complete protease inhibitors (Millipore Sigma) and incubated on ice for 5 min.Lysates were further disrupted with sonication 3 times at 30% output (Vibra-Cell, SONICS) and centrifuged at 16 000 g for 15 min at 4 • C. Supernatants were flash frozen in liquid nitrogen and stored at −80 • C. For IP, lysates were mixed with antibodies against MIWI or MILI and incubated at 4 • C for 2 h.Lysis buffer equilibrated Protein G Dynabeads (Invitrogen) were added to the mix and incubated for 90 min at 4 • C. Beads were washed 4 times in lysis buffer for 5 min.For WB, protein lysate / eluates were heated in SDS-loading buffer for 12 min at 70 • C, resolved on 4-12% NuPAGE Bis-Tris gels (Invitrogen), blotted to nitrocellulose membranes (Invitrogen), blocked with 5% nonfat milk and incubated with primary antibodies overnight.MVH antibody was produced by immunizing rabbits with synthetic peptide SSQAPNPVDDESWD conjugated to KLH protein via an amino-terminal cysteine, followed by affinity purification of sera over columns containing the immobilized peptide (Genscript).Quantitation of protein levels from WBs was performed using ImageJ ( https:// imagej.net/ij/ index.html ) and beta-tubulin was used for normalization.All antibodies used for IP and WB are listed in Supplementary Table S13 .

piRNA library preparation
Radiolabeled piRNAs were resolved and gel-purified from 8 M urea 15% polyacrylamide gels (PAGE).Eluted RNA was ligated to miRCat 3 Linker-1 (IDT) modified with eight extra random nucleotides at the 5 -end.Ligation was performed in 25% PEG 8000 ( 42 ) by T4 RNA Ligase 2 Truncated K227Q (NEB) for 8 h at 16 • C. Ligated RNA was PAGE purified and reverse-transcribed with a 5 phosphorylated long primer containing 3 and 5 adaptor complementary sequences ( 43 ).Primer and adaptor ligated-RNA were heated at 65 • C for 10 min and cooled down at room temperature for 5 min.cDNA was produced by AffinityScript (Agilent) in the presence of [ α-32 P] ATP (10mCi / ml, 3000 Ci / mmol; Perkin Elmer, BLU512H250UC).After P AGE purification, cDNA was circularized by CircLigase I (LGC Biosearch Technologies) and amplified with 25 PCR cycles by Phusion High-Fidelity DNA polymerase (NEB) and adaptor specific primers carrying Illumina p5 and p7 flow cell binding sequences, followed by Illumina 150 Paired-End sequencing in biological replicates.The oligonucleotides used for library construction are listed in Supplementary Table S14 .
Ribo-seq and RNA-seq P24 testes were used for preparation of all libraries with the ligation-free method as described by the Sims lab ( 44 ), using the SMARTer smRNA kit for Illumina (Takara), with few modifications.For Ribo-Seq, testes lysates were treated with RNAse I and subjected to monosome isolation by S400 spin column centrifugation.Ribosome footprints were purified with 15% UREA-PAGE.For RNA-Seq, total RNA was used after ribodepletion (Ribominus eukaryotic kit v2; Ther-moFisher) and fragmentation (25 min at 94 • C in 2 × T4 PNK reaction buffer from NEB-140 mM Tris-HCl pH 7.6, 20 mM MgCl 2 , 10 mM DTT) to obtain RNA fragments of similar size to ribosome footprints.RNA fragments (either ribosome protected footprints or RNA-Seq) were in vitro polyadenylated and reverse-transcribed with an oligo(dT) primer and a template switching oligo, followed by 12 cycles of PCR amplification with adapter specific primers carrying Illumina p5 and p7 flow cell binding sequences, followed by Illumina 50 Single-End sequencing in biological replicates.Sequencing and library preparation were performed at TB-SEQ.

RNA extraction, reverse transcription and qPCR
Total RNA was extracted from P24 testes with TRIzol reagent (Invitrogen) and treated with RQ1 DNAse (Promega).2 μg total RNA was reverse-transcribed by Superscript III with random hexamers (Invitrogen).cDNAs were amplified using PowerUp SYBR-green mix (Applied Biosystems) and the primers listed on Supplementary Table S15 .Assays were run on StepOnePlus Real-Time PCR system (Applied Biosystems).Tcp1 mRNA was used for input and differentiation stage normalization.Results were calculated from biological triplicates.

Degradome-Seq library preparation
For Degradome-Seq (PARE-seq), polyadenylated RNA was purified with oligo(dT) beads (NEB) from 15 μg of total RNA extracted from P24 testes.Poly(A) RNA with 5 phosphates was ligated to a 5 RNA adapter (CU AC ACGACGCUCUUC-CGAUCUNNN) by T4 RNA ligase I and reverse-transcribed with random hexamers.Libraries were prepared using the NEBNext Ultra II RNA Library Prep Kit (NEB) followed by Illumina 150 Paired-End sequencing in biological replicates.Library preparation and sequencing were performed at CD Genomics.

Data analysis
The visualizations and statistical analyses were performed in R (v3.6.3).All software tools were installed and run using Conda (v4.11.0) environments.Ensembl genome reference (GRCm38), Ensembl gene annotation, and UCSC Re-peatMasker track were used for all analyses.Sequencing batch effects were removed from expression calculation where applicable.
piRNA distribution piRNA read length was summarized by bbmap (v38.67).Nucleotide composition and genomic origin were determined by piPipes (commit c93bde3).For 3 end comparisons of piR-NAs, mapped MILI-bound and MIWI-bound piRNAs with the same starting genomic position (same 5 end) with read lengths between 24 nt and 40 nt were used.The median length of each piRNA (per 5 start position) was determined and the difference in median lengths between Miwi + / RK and Miwi RK / RK were calculated and plotted.To calculate 5 -5 end distances, piRNA reads had to map on the opposite strand within a 30 nt window.The mapped 5 positions and the distances were determined by bedtools (v2.29.0).Only uniquely mapped reads were considered.The z -score for 10 nt overlap (ping-pong amplification signature) was calculated with:

Gene expression and ribosome occupancy
Transcript expression and ribosome occupancy were estimated with Salmon (v0.14.1).Gene expression was calculated by summarizing transcript-to-gene expression with tximport (v1.12.1) and edgeR (v3.26.0) was used to calculate the differential expression.Differential ribosome occupancy was calculated by edgeR with the formula ∼batch + condition + type + condition:type where: batch is the sequencing batch information; condition is Miwi + / RK (Het) or Miwi RK / RK (RK); type is RNA or RIBO; condition:type is the interaction between the conditions and types.Genes with adj.Pvalue < 0.05 and fold change ≥2 were considered as differentially expressed or with differential ribosome occupancy.

piRNA expression and targeting
We used the first 25 nucleotides downstream from the first mapped position from the genomic reference as piRNArepresentative sequences.piRNA expression was calculated by summing up abundances of all identical piRNArepresentatives. edgeR (v3.26.0) was used to calculate the differential expression.piRNAs with adj.P -value < 0.05 and fold change > 0 were considered as differentially expressed.piRNA targeting was predicted with GTBuster (commit 6697717).The default settings were modified to keep only results, where targeting piRNA is complementary to the targeted transcript with mismatch allowed at the first position, minimum of 14 matches between nucleotides 7-25, and degradome-seq supported cut between 10-11 nt of the piRNA sequence.Only Miwi-expressed transcripts ( ≥100 RNA-Seq reads in at least one sample) were used to estimate the theoretical targeting.
Predicted cuts were overlapped with the UCSC RepeatMasker track by bedtools (v2.29.0).Degradome-Seq supported cuts had to be present in both biological replicates to be considered in subsequent steps.The actual targeted and regulated genes were determined by combing the predicted targeting with gene (adj.P -value < 0.1, fold change > 0) and piRNA (adj.P -value < 0.25, fold change < 0) differential expression.

MIWI-NTRs are essential for fertility and spermiogenesis
To define the molecular function of MIWI's NTRs, we used genome editing of the endogenous Miwi locus to replace the six arginine residues (R4, R6, R8, R10, R12 and R14), which are subjected to symmetrical dimethylation, with lysines to generate the Miwi RK allele (Figure 1 A and Supplementary Figure S1 A-D).Unlike wild-type mice and heterozygous littermates, Miwi RK / RK males are infertile (Figure 1 B, Supplementary Table S1 ), with 35% smaller testes (Figure 1 C, Supplementary Table S2 ) carrying no sperm in the seminiferous tubules or epididymis (Figure 1 D), due to a spermatid elongation failure phenotype and spermiogenesis arrest at steps 8-9 (Figure 1 E).The MIWI-RK protein does not show any toxic or gain of function properties as it does not affect fertility when expressed along with wild-type MIWI in heterozygous Miwi + / RK males.The first wave of spermiogenesis in mice is synchronized, with round spermatids appearing at postnatal day 20 (P20) and spermatid elongation commencing around

MIWI-NTRs sustain amplification of pachytene cluster piRNAs and transposon control
To examine MIWI methylation, we probed MIWI immunoprecipitates (IP) from testes on WB with SYM11, an antibody that specifically recognizes sDMAs.As shown in Figure 2 A, SYM11 reactivity of MIWI is lost in Miwi RK / RK mice , con-sistent with R4, R6, R8, R10, R12 and R14 being the main arginines in MIWI that are subjected to symmetric dimethylation.Next, we examined with co-IP and WB assays whether interactions between MIWI-RK and TDRDs were altered.As shown in Figure 2 B, MIWI-RK fails to interact with TDRD5 and TDRKH.Since MIWI-RK abolishes interactions with TDRDs that are essential for pachytene piRNA biogenesis, we hypothesized that piRNA populations in Miwi RK / RK mice would be affected.One possibility is that MIWI-RK would be devoid of most piRNAs since it does not interact with TDRKH, whose postnatal deletion leads to a near collapse of MIWI piRNAs ( 38 ).Another possibility is that the piRNA composition and genomic origin would be altered since MIWI-RK does not interact with TDRD5, which is required for piRNA generation from the entirety of pachytene piRNA precursors ( 37 ).We immunoprecipitated MILI and MIWI from Miwi + / RK and Miwi RK / RK lysates, isolated bound piRNAs and generated sequencing libraries (biological duplicates).Surprisingly, MIWI and MILI proteins are loaded with piRNAs in both genotypes (Figure 2 C), without any major differences in nucleotide composition ( Supplementary Figure S2 A), piRNA sizes ( Supplementary Figure S2 B,C), genomic origin ( Supplementary Figures S3 -S6 ), or overall differential expression ( Supplementary Figure S7 ) indicating that MIWI-RK does not overtly affect piRNA biogenesis, precursor-transcript processing or piRNA loading.Ultrastructural analysis of spermatocytes in Miwi + / RK and Miwi RK / RK shows that IMC is indistinguishable between the two genotypes (Figure 2 D).This is further indication that MIWI-RK does not have a major impact in piRNA biogenesis.Next, we examined whether piRNA amplification ('pingpong') was affected by MIWI-RK by calculating the fraction of piRNAs in ping-pong pairs ( Supplementary Tables S3 -S8 ) and by plotting the 5 -5 distance of MILI-and MIWI-bound piRNAs in Miwi + / RK and Miwi RK / RK .For MILI piRNAs, the 5 -5 position analysis shows a peak at distance 10, with slightly increased z -scores in Miwi RK / RK , consistent with largely intact MILI piRNA ping-pong (Figure 2 E, Supplementary Figure S8 A).In contrast, ping-pong zscore values between MIWI bound piRNAs are reduced in Miwi RK / RK , indicative of attenuated piRNA amplification by MIWI-RK ( z -score of 13.86 in Miwi RK / RK versus 17.36 in Miwi + / RK for piRNAs mapping to pachytene clusters, Figure 2 F; z -score of 13.15 in Miwi RK / RK versus 16.19 in Miwi + / RK for piRNAs mapping to genes, Supplementary Figure S8 B). z -scores were very similar for heterotypic, MILI-MIWI pingpong between the two genotypes ( Supplementary Figure S9 ).Such attenuation likely underlies the reduction of MIWI-RK protein levels in Miwi RK / RK (Figure 1 F), since cleavage of piRNA precursors by piRNAs, also promotes piRNA biogenesis ( 1 ).To examine the impact of attenuated ping-pong in transposons, we analyzed the levels of individual transposable elements (TEs) by performing RNA-Seq from total RNA isolated from Miwi + / RK and Miwi RK / RK testes.While the levels of the vast majority of TEs are not altered between the two genotypes, we find that five times as many TEs are upregulated than downregulated in Miwi RK / RK (133 upregulated, 28 downregulated, Figure 2 G and Supplementary Table S9 , adj.P -value < 0.05, fold change ≥ 2), indicating that attenuated piRNA amplification decreases clearance of select TEs in Miwi RK / RK mice.

MIWI-RK impacts the mRNA transcriptome but not the translatome
To investigate the possible impact of MIWI-RK on the expression and translation of mRNAs, we performed, in biological triplicates, ribosome profiling (Ribo-Seq) from Miwi + / RK and Miwi RK / RK testes and analyzed the libraries along with concurrent RNA-Seq.Although the expression levels of most transcripts are not altered between the two genotypes, we find that 180 (116 protein-coding) genes are upregulated and 71 (38 protein-coding) genes are downregulated in Miwi RK / RK (Figure 3 A and Supplementary Table S10 ; adj.P -value < 0.05, fold change ≥ 2).We find that the ribosome occupancy is essentially identical in both genotypes, with changes in only a handful of genes.Specifically, 14 transcripts (10 genes) were more occupied and 15 (12 genes) less occupied (adj.Pvalue < 0.05, fold change ≥ 2), indicating that mRNA translation is not grossly impacted in Miwi RK / RK mice (Figure 3 B and Supplementary Table S11 ).

MIWI-NTRs sustain piRNAs that cleave and destabilize select mRNAs essential for spermatogenesis
Since piRNA amplification is attenuated by MIWI-RK, we hypothesized that some of the upregulated transcripts in Miwi RK / RK could be directly targeted and cleaved by piRNAs with decreased levels in the mutant.To investigate our hypothesis, we first predicted RNA targets that are potentially cleaved by piRNAs.We used parameters that considered the requirement for both extended stretches of complementarity between the piRNA and its target, and tolerance for mismatches ( 9 ,47-49 ).We find that of the 45 029 transcripts expressed at P24 testes, 31745 ( ∼70%) are theoretically targeted by 46 494 out of 226 198 unique MIWI-bound piRNAs.The meta-transcript distribution of predicted target sites in meta-mRNA is heavily skewed towards the 3 Untranslated Region (3 -UTR, Figure 4 A).
PIWI nucleases cleave the phosphodiester bond of target RNAs across from the 10th and 11th nucleotide of their bound piRNA, generating a decay fragment that has a 5 phosphate group, which can be captured and sequenced with Degradome-Seq.We performed Degradome-Seq from poly(A) RNA isolated from Miwi + / RK and Miwi RK / RK testes (biological duplicates).To identify which of the predicted target sites were cleaved by piRNAs, we analyzed the position of the first nucleotide of decay intermediates captured with Degradome-Seq in relationship with piRNA-target RNA pairs.We find that the targeting of 686 transcripts (678 genes out of 31 745; 2%) by 662 unique, MIWI-bound piRNAs, is supported by Degradome-Seq (meta-transcript distribution of target sites with Degradome-Seq support in the two genotypes is shown in Figure 4 B).More than half (374 out of 662) of the unique MIWI-bound piRNAs, originate from repeat-related loci in either sense or antisense orientation to the repeat annotation.The predicted piRNA target sites are primarily in 3 UTRs (764 target sites), while Coding Sequence (CDS) and 5 Untranslated Region (5 UTR) are much less favored (170 and 29 target sites, respectively).
Next, we identified MIWI-bound piRNAs whose expression is lower in Miwi RK / RK , compared to Miwi + / RK , and with corresponding mRNA targets upregulated.We find that of 678 genes targeted by piRNAs, 17 are upregulated (adj.Pvalue < 0.1, fold change > 0) in Miwi RK / RK with corresponding downregulation (adj.P -value < 0.25, fold change > 0) of 19 targeting, MIWI-bound piRNAs (Figure 4 C).Notably, the sequences of half of the piRNA-target RNA pairs are derived from repeat elements while few piRNAs targeted piRNA cluster transcripts (Figure 4 C).Strikingly, 3 of the 17 upregulated genes in Miwi RK / RK along with the corresponding downregulated targeting piRNAs, are the same as those identified in the pi6 ( Dnajc3 , Kctd7 ) and pi18 ( Golga2 ) knockouts ( 18 ,19 ) (Figure 4 C).Another two ( Stambp2 and Tdrd1 ) were previously shown to be upregulated in Miwinull mice ( 13 ).GOLGA2 (GM130) is a Golgi matrix protein essential for acrosome formation ( 50 ).Ultrastructural examination of Miwi + / RK and Miwi RK / RK testes shows that the acrosome is intact in Miwi + / RK spermatids, with normal formation of acroplaxome, acrosomal granule and acrosomal vesicle ( Supplementary Figure S10 ).However, acrosome is not formed in Miwi RK / RK spermatids, which in most cases show a mis-oriented Golgi apparatus that does not form an acrosome or less frequently multiple, small fragmented acrosomal granules that do not coalesce to form an acrosome ( Supplementary Figure S10 ).The phenotype of Miwi RK / RK is more similar to that of Miwi -null ( 5 , 12 , 13 ) and much more severe than that seen with loss of individual piRNA loci ( 18 ,19 ), consistent with the larger transcriptome alterations found in Miwi RK / RK mice.

MIWI-NTRs are required for TDRD6 interaction, chromatoid body compaction and the stability of key spermiogenic transcripts
Next, we tested with co-IP and WB assays whether the binding between MIWI-RK and TDRD6 was abolished.As shown in Figure 5 A, MIWI-RK fails to interact with TDRD6.We examined by immunofluorescence (IF) the localization of MIWI and TDRD6 in testes from heterozygous and homozygous Miwi RK mice.As shown in Figure 5 B, there is similar localization and CB formation, in both genotypes.However, ultrastructural analyses show that CB is less compact, less electron dense and often fragmented in Miwi RK / RK round spermatids (Figure 5 C), indicating that MIWI-TDRD6 interaction is required for CB compaction.The CB defect in Miwi RK / RK is similar to what is seen in deletion mutants of Miwi or Tdrd6 ( 5 , 20 , 23 ).
Among the 38 downregulated genes in Miwi RK / RK ( Supplementary Table S10 ) many are essential spermiogenic genes that were previously shown to be downregulated in Miwi -null mice ( 5 ,12 ).These include genes coding for nuclear transition proteins ( Tnp2) and protamines ( Prm1 , Prm2 ), which replace the histones during spermatid elongation to form the sperm nucleus ( 51 ,46 ).To examine whether TDRD6 has similar impact on the transcriptome, we analyzed previ- ously published RNA-Seq data from sorted P26 spermatids from Tdrd6 -/ + and Tdrd6 -/ -mice ( 52 ) and compared them to Miwi RK / RK .We find 6 downregulated genes (adj.Pvalue < 0.05, fold change ≥ 2) that are common to both Tdrd6 -null and Miwi RK / RK mice, including Tnp2 , Prm1 and Prm2 ( Supplementary Table S12 ).To examine whether the reduced transcript levels in Miwi RK / RK were secondary to reduced production or destabilization, we performed RT-qPCR, in biological and technical triplicates, and utilized the Tcp1 transcript for input and differentiation stage normalization ( 53 ).We find that mRNA levels for Prm1 , Prm2 and Tnp2 are significantly reduced in Miwi RK / RK compared to in Miwi + / RK , while their pre-mRNA levels are either unchanged or slightly upregulated (Figure 5 D), indicating mRNA destabilization of key spermiogenic transcripts by MIWI-RK.Collectively, these findings demonstrate that downregulation of spermiogenic transcripts required for chromatin remodeling and sperm nuclear compaction is a shared molecular phenotype of mice null for either Miwi or Tdrd6 , as well as Miwi RK / RK mice, which are unable to form MIWI-TDRD6 complexes.

Discussion
Prior genetic investigations of MIWI and TDRDs relied on gene knockouts and illuminated key aspects of piRNA biology but could not directly address the role of MIWI-TDRD interactions in piRNA biogenesis and function, in vivo .Here, with the separation-of-function Miwi RK allele we specifically uncouple MIWI from TDRDs, while preserving expression of all TDRDs and functions associated with all other PIWI domains.A proposed model that emerges from our findings is shown in Figure 6 and is discussed below.
The Miwi RK allele illuminates functions of MIWI in piRNA amplification, biogenesis and piRNA targeting We find that MIWI-RK attenuates piRNA amplification, leading to reduced transposon clearance and piRNA biogenesis.By disrupting the interaction of MIWI with TDRDs that are involved in piRNA biogenesis, we unveil the few functional piRNAs that target specific mRNAs for destabilization.Amongst them are previously reported gene targets ( 13 ) and piRNAs from the pi6 and pi18 loci, recently shown to be essential for mouse fertility ( 18 ,19 ).We also uncover targeting by 'nonfunctional' piRNAs that appear to cleave mRNAs without impacting their stability or translation.We find that the sequences of most piRNAs that cleave mRNAs (both functional and 'nonfunctional') are related to repeat elements.In turn, these piRNAs target repeat element sequences embedded in mRNAs, typically in their 3 -UTRs.It is not clear why some mRNAs cleaved by piRNAs are not destabilized.We favor the idea that such cleavages are infrequent events that affect only a small fraction of mRNA molecules.This could be due to a number of factors, such as low levels of the targeting piRNA and / or large number of target sites distributed amongst TE transcripts and the targeted mRNAs, diluting the impact of targeting piRNAs on the protein-coding transcriptome.We find that the vast majority of pachytene piRNAs are not involved in gene regulation, consistent with our previous hypothesis ( 12 ) and with more recent reports ( 54 , 18 , 19 ).Our findings support the primacy of MIWI piRNA amplification for post-meiotic clearance of TE transcripts ( 9 ).They also support the notion that the vast majority of pachytene piRNAs act as self-serving genetic elements that rely for propagation on few functional piRNAs ( 18 ,19 ).We propose that production of these functional piRNAs depends on MIWI piRNA amplification, which in turn is regulated by MIWI-NTRs.Since piRNA amplification involves mostly repeatrelated elements whose sequences are divergent in different mammalian species, the functional piRNAs that arise are not conserved.Our proposition is further supported by the demonstration that RNF17 / TDRD4 ( 33 ) is a negative regulator of meiotic piRNA amplification ( 36 ).In the absence of RNF17, hyperactive piRNA amplification leads to inappropriate targeting and downregulation of multiple proteincoding genes ( 36 ) and spermiogenic arrest at the round spermatid stage ( 33 ,36 ) (Figure 6 A).
The Miwi RK allele illuminates the primacy of MIWI-TDRD6 interactions in chromatoid body compaction and spermiogenic mRNP stability Tnp2, Prm1 , Prm2 whose translation is delayed until spermatid elongation ( 12 ).By analyzing published RNA-Seq data from sorted P26 spermatids of Tdrd6 -/ + and Tdrd6 -/ -mice ( 52 ), we find that the same genes are downregulated in the absence of TDRD6.We show that the interaction of MIWI with TDRD6 is required for CB compaction.Furthermore, we solidify the role of MIWI and its NTRs in stabilization of these key spermiogenic mRNAs that drive histone replacement of spermatid nuclei.The molecular mechanism that underlies spermiogenic mRNA stabilization remains to be determined but likely involves a non-cleaving, mRNA binding mode for MIWI.We previously reported that MIWI binds directly to translationally repressed spermiogenic mRNAs (by CLIP-Seq and CIMS -crosslinked induced mutation siteanalyses) and proposed that it does so without using piR-NAs ( 12 ).It is also possible that MIWI piRNAs bind spermiogenic mRNAs with partial complementarity that does not lead to mRNA cleavage.We have shown that such a binding mode occurs in Drosophila melanogaster , where Aub-piRNAs bind mRNAs with partial complementarity in the germ plasm and along with Tudor protein form germ granules ( 55 ).Further studies are required to identify the precise molecular function of MIWI in spermiogenic mRNP assembly and stabilization.

Parallels between the functions of Mus musculus MIWI / TDRDs and Drosophila melanogaster Aub / TDRDs
Our findings draw parallels between the functions of mouse MIWI / TDRDs and D. melanogaster Aub / TDRDs.We recently examined, genetically, the role of Aub-NTRs and their methylation status in piRNA biogenesis, amplification and formation of germ granules ( 56 ).The latter consist of mRNPs that contain mRNAs, bound directly by Aub ( 55 ), and also contain Tudor ( 55 , 57 , 58 ), Vasa (59)(60)(61) and other RNAbinding Proteins ( 62 ).Germ granules assemble at the posterior of D. melanogaster oocytes and their mRNAs are essential for primordial germ cell specification in the developing embryos ( 62 ).By genetic deletion or germ cell specific knockdown of Drosophila Prmt5 , we found that germ granules never formed as Aub with unmethylated NTRs could not interact with Tudor ( 56 ).piRNA biogenesis, amplification and transposon control were intact in flies expressing endogenous Aub with unmethylated NTRs ( 56 ).However, by mutating Aub-NTRs to lysines, we found that germ granule formation, piRNA amplification and transposon control all collapsed, as the Aub RK mutant protein was unable to interact not only with Tudor but also with TDRDs involved in piRNA amplification ( 56 ).Primary piRNA biogenesis was not impaired by Aub RK ( 56 ).We think that similar functional interactions take place in mice between MIWI and TDRDs.MIWI-RK, by disrupting interactions with TDRDs involved in piRNA biogenesis and with TDRD6, leads to attenuation of piRNA amplification and biogenesis, TE transcript upregulation, CB decompaction and spermiogenic mRNA destabilization.We speculate that sDMAs of MIWI-NTRs and the valency of their interactions with the multiple eTUD domains of TDRD6, drive functions relating to CB compaction and spermiogenic mRNP formation.
While our manuscript was under revision, Wei et al. reported the generation of a Miwi RK mouse mutant, with spermiogenesis arrest at step 8 ( 63 ), similar to what we find.Wei et al. characterized their allele in a Miwi null background ( Miwi RK / -) and found that MIWI-RK protein is reduced with concomitant reduction of their bound pachytene piRNAs, which are qualitatively similar to those bound by wild-type MIWI ( 63 ).We also find that Miwi RK does not overtly affect pachytene piRNA biogenesis or piRNA loading but we do not observe the large reduction in the levels of MIWI-RK protein or piRNAs, since we characterize mice that are homozygous for the mutant allele ( Miwi RK / RK ).Importantly, by performing deep characterization of our Miwi RK mice, with RNA-Seq, Ribo-Seq, Degradome-Seq, piRNA-Seq and extensive bioinformatic analyses, we provide molecular explanations for the role of MIWI NTRs in spermiogenesis and pachytene piRNA function.
In summary, we find that a central molecular mechanism by which PIWI proteins are directed and utilized by distinct cellular pathways, resides in the NTR domain.This mechanism is essential for the integrity and continuity of the germline.

Figure 4 .
Figure 4. MIWI-NTRs sustain piRNAs that clea v e and destabilize select mRNAs essential for spermiogenesis.Metatranscript distribution of predicted, piRNA-mediated clea v ages on mRNA targets, without ('theoretical', A ), or with degradome-Seq support ( B ). ( C ) Sequence and genomic location of piRNAs found to cleave and destabilize the listed mRNAs at the indicated coordinates.piRNAs and clea v ed RNA fragments map to the f ollo wing categories (green bo x es): Cl: piRNA cluster; S: sense-aligning repeat sequence; AS: antisense-aligning repeat sequence; E: e x onic sequence.Arrows indicate changes in the expression of piRNAs (down) and mRNA targets (up) in Miwi RK / RK compared to Miwi + / RK .