The piRNA Pathway: Guardian of the Genome –A Fly’s Perspective

Abstract

Throughout the eukaryotic lineage, small RNA silencing pathways protect the genome against the deleterious influence of selfish genetic elements such as transposons. In animals an elaborate small RNA pathway centered on PIWI proteins and their interacting piRNAs silences transposons within the germline, which transmits the genome to future generations. In contrast to other small RNA silencing pathways, we entirely lack a mechanistic understanding of this genome defense system. However, genetic and molecular work over the last ten years has uncovered a fascinating framework of this pathway that is conserved from sponges to mammals. This review discusses our current understanding of the piRNA pathway in Drosophila with an emphasis on origin and biogenesis of piRNAs.

Eukaryotic genomes harbor a variety of selfish genetic elements, stretches of DNA that gain a transmission advantage relative to the rest of the genome, while not increasing the organism’s fitness¹. The best-understood and most widespread selfish elements are mobile elements called transposons². The success of these “genome parasites” rests on their ability to multiply within the genome by transposition to new sites. This ultimately affects host fitness due to insertional mutagenesis and ectopic chromosomal recombination. Throughout the eukaryotic lineage, the threat posed by transposons is met by host defense systems that selectively silence them. Though early genetic studies have illustrated the existence of such defense-systems^3,4, their molecular nature remained mysterious for a long time. This changed abruptly when the concept of small RNA pathways, which govern RNA mediated silencing phenomena was discovered^5–8. Over the last ten years it has become increasingly evident that small RNA silencing pathways protect the genomes of plants, fungi and animals against transposons and other selfish elements^9,10.

In this review, we discuss a small RNA silencing pathway that is selectively active in animal gonads where it safeguards the genome of reproductive cells against transposons. This so-called piRNA pathway centers on PIWI family proteins and their bound PIWI interacting RNAs (piRNAs). Our article focuses on the piRNA pathway in the Drosophila ovary, where a long history of genetic research combined with recent small RNA centered studies has revealed the conceptual framework of this genome surveillance system that is conserved from sponges to mammals.

We first outline the common logic of small RNA silencing pathways. We then describe the architecture of the Drosophila ovary as it allows to conceptually separate two distinct but related piRNA pathway modules. Separate discussions on these two modules constitute the major part of the review.

Concepts of small RNA silencing pathways

Common to all small RNA pathways is a silencing machine called the RNA induced silencing complex (RISC). Its central components are an Argonaute family protein and a bound small RNA¹¹. Via complementary base pairing the small RNA guides RISC to cellular target RNAs, which typically results in target silencing. The remarkable elegance of small RNA pathways is based on their inherent simplicity. In one or the other form, an mRNA is a key intermediate of all gene expression programs. Thus, loading an Argonaute protein with a small RNA complementary to the target gene allows the inhibition of essentially every cellular process^12,13.

Argonaute proteins have diversified during evolution and in most animals three small RNA pathways can be distinguished. These are the ubiquitous microRNA and small interfering (siRNA) pathways and the generally germline specific piRNA pathway (Box 1).

BOX1. Small RNA pathways.

Genetic studies have identified several concepts of small RNA mediated regulation years before RNA interference and related small RNA pathways were described²⁸,⁸⁵,⁸⁶. Nevertheless, the milestones in the field were the discoveries of dsRNA as trigger for RNA interference and the identification of small RNAs and Argonaute proteins as the key components of all small RNA pathways⁵–⁸. For the first time, scientists had a molecular entry point into a world of novel and diverse biology. What followed was an explosion in our knowledge about mechanisms and functions of a still growing number of small RNA pathways.

Evolution has shaped a diverse array of pathways from the common principle of target repression via small complementary RNA guides. This is best illustrated by the radiation of Argonaute proteins, the universal binding partners for small RNAs, which in many cases are able to cleave (slice) the target upon successful recognition⁸⁷,⁸⁸.

In most animals, two classes of Argonaute proteins, the AGO subfamily and the PIWI subfamily can be distinguished (a third subfamily, the so-called WAGO proteins, has been identified only in nematodes⁸⁹). AGO proteins are expressed ubiquitously and are loaded with microRNAs and endogenous siRNAs in response to specific dsRNA triggers. While microRNAs guide the regulation of endogenous gene expression programs⁹⁰, siRNAs are mostly involved in the suppression of foreign gene expression¹³, be it from viruses of from selfish genetic elements. In flies, the siRNA pathway is much more elaborate than in mammals, presumably as insects lack the sophisticated adaptive immune system.

Most animals possess two or three PIWI family proteins that are typically expressed in gonads. Though flies and mice express both three PIWI proteins, pair wise orthologies cannot be determined. In fact, the three Drosophila PIWI clade proteins, Piwi, Aubergine and AGO3 are more related to each other than to the mouse PIWI members MILI, MIWI and MIWI2. This might suggest that PIWI proteins radiated within lineages from a single ancestral protein. Mutations in PIWI family proteins lead to sterility and severe defects in gametogenesis in all animals examined so far¹⁰,¹⁴.

A key advantage of small RNA pathways in the defense against foreign genetic elements is that the target sequence can also act as the trigger for small RNA biogenesis. Small RNAs are thereby inevitably coupled to their target, even if target sequences evolve rapidly. In most animals, the siRNA and piRNA pathways implement this principle. Within the siRNA pathway, intra- or inter-molecular double stranded RNA (dsRNA) originating from transposons or viral transcripts triggers siRNA production via Dicer. Loaded into RISC, these siRNAs guide target silencing in trans, and provide a protective layer in somatic cells^12,13 (Box 2). The real battle, however, takes place in the germline, where transposons are particularly active due to their predominantly vertical transmission strategy. Here, the piRNA pathway silences selfish elements to ensure genetic stability across generations^10,14. piRNA biology differs considerably from that of other small RNA pathways and almost nothing is known about piRNA biogenesis and their mode of action.

BOX2. Transposon silencing by endogenous siRNAs.

In flies, deep sequencing of small RNAs bound to AGO2 from somatic and germline tissues identified a large fraction of endogenous siRNAs (endo-siRNAs) with sequences corresponding to transposons and other genomic repeats³³,³⁴,³⁶,⁹¹. Presumably, dsRNA from sense and antisense transcripts triggers their production. In ovaries, endo-siRNA profiles therefore overlap with those of piRNAs, yet they lack a similar antisense bias. Significant de-repression of several transposons at the RNA level has been observed in ovaries and flies mutant for AGO2, Dicer-2 or Loquacious, the three key factors for the endo-siRNA pathway. Thus, in gonads piRNAs and siRNAs collaborate in transposon silencing, though the extent of repression (at least for several elements) appears to be much higher for the piRNA pathway²⁶,³². The Penelope element of Drosophila virilis, however, is largely controlled via endo-siRNAs with only very few piRNAs targeting this element being identified⁹². In non-gonadal tissues, the endo-siRNA pathway appears to be the only line of defense against transposons, though additional repression at the chromatin level cannot be excluded.

In mammals, endo-siRNAs have so far been only identified in oocytes, where they have an important role in transposon control⁹³,⁹⁴. Also studies in C. elegans indicate a much more pronounced role of endo-siRNAs in the defense against transposons ⁹⁵–⁹⁷. Here, however the nemotode specific WAGO clade of Argonaute proteins makes direct comparisons to the above described siRNA pathway increasingly difficult.

In flies and mice, some endo-siRNAs are also generated from piRNA clusters³⁴,³⁵. This might simply reflect the ability of piRNA cluster transcripts to form low levels of dsRNA with complementary transposon transcripts. A significant mechanistic connection between the two pathways seems unlikely as piRNA pathway mutants show little impact on siRNA populations and siRNA pathway mutants do not affect piRNA pools³¹,³²,³⁴.

The Drosophila ovary: Evidence for distinct piRNA modules

The Drosophila ovary consists of germline cells and somatic support cells (mostly follicle cells) that have key roles in maintaining and protecting the germline cells (Fig. 1A). Within the germline, nurse cells and oocyte share a common syncytial cytoplasm. Intricate connections also exist between germline and somatic cells, e.g. via exchange of developmental signals or nutrient flow into the germline. This exposes the oocyte genome to an internal and external threat: On the one hand, several transposons are highly active in germline cells and exploit cellular machineries to maximize access to the oocyte genome¹⁵. On the other hand, several retro-elements from the gypsy family form viral particles in somatic support cells that invade the germline, presumably via cellular transport vesicles ^16–19. In both cell types, the piRNA pathway is the major line of defense against transposons. How silencing is achieved, however, differs significantly. In fact, while germline cells express three PIWI family members (Piwi, Aubergine (Aub) and AGO3), somatic support cells express exclusively Piwi^20–22. In the following, we first describe the linear piRNA module that acts in ovarian somatic support cells. We then build on these concepts to introduce the more complex scenario in the germline, where a piRNA amplification module based on PIWI proteins is active.

Figure 1. A primary piRNA pathway is active in somatic cells of the Drosophila ovary.

A) The Drosophila oocyte is in direct contact with germline-derived cells (beige) and is surrounded by cells of somatic origin (green). This cartoon depicts an ovariole, the functional unit of the ovary (reproduced with kind permission from A. Spradling ¹¹⁰). Development proceeds from left (germarium) to right (mature egg). In the germarium, Germline Stem Cells (GSCs) divide asymmetrically into GSCs and differentiating cystoblasts. Four mitotic cystoblast cell divisions produce 15 nurse cells and an oocyte that remain connected by cytoplasmic bridges. Each of these germline cell clusters is surrounded by an epithelium of somatic follicle cells (green) to form an egg chamber that continuously grows until the oocyte matures into an egg. Follicle cells finally undergo apoptosis after depositing the eggshell. The deposited egg therefore lacks somatic cells. B) Shown is a schematic representation of the somatic piRNA pathway (primary piRNA module). For illustrative purposes, piRNA source and target loci from the X-chromosome (drawn at the top) are displayed. Colored boxes summarize primary piRNA biogenesis from piRNA clusters (yellow) and from 3’UTRs of protein coding genes (green). The blue box shows expression and silencing of ZAM, a prototypical LTR-retrotransposon, active in follicle cells. (yellow box) The piRNA clusters 20A and flamenco are located at the boundary between euchromatin and heterochromatin. Both contain almost exclusively transposon fragments oriented antisense to the unidirectional promoter. piRNA cluster transcripts (red) therefore give rise to antisense piRNAs. Unknown mechanisms parse piRNA cluster transcripts into shorter fragments that might enter Piwi. At this step, Piwi could preferentially select precursors with a 5’ Uridine (1U). Subsequently, the 3’ tail of Piwi bound RNAs is trimmed and 2’OH-methylated to generate mature piRNAs. (blue box) The sequence of mature piRNAs defines their target: Displayed is an active copy of the ZAM LTR-retrotransposon and its sense transcript (blue box). The green box summarizes piRNA biogenesis from genes (here diminuitive). The spliced dm transcript with 5’UTR, coding sequence and 3’UTR is shown. Mature dm mRNAs are either translated into Myc or act as piRNA precursors. piRNAs are preferentially processed from 3’UTR sequences, presumably by a similar mechanism as for piRNA cluster transcripts. Genic piRNAs are in sense orientation to the host gene and their targets (if any) remain to be identified.

A linear piRNA module in somatic gonadal cells

piRNA biology in the ovary-soma exhibits an overall simple architecture (Fig. 1B). All somatic support cells express Piwi, the only nuclear PIWI family protein in flies^23,24. Piwi binds a spectrum of predominantly transposon derived piRNAs^20,25–27 and silences transposon expression by an unknown mechanism. The only other factor with an understood function is the X-chromosomal flamenco locus that serves as a major source for piRNAs^20,28–31. To a large extent, deep-sequencing of piRNA populations provided our current insight into somatic piRNA biogenesis and their target spectrum and biological roles.

Somatic piRNAs and their origin

The purest datasets of somatic piRNAs were obtained from an ovarian somatic sheet cell line (OSS cells)^21,22. We will use this dataset to illustrate key features of somatic piRNAs. We note that populations of Piwi bound piRNAs from entire ovaries and the population of piRNAs that is selectively found in ovaries but not in eggs (during final stages of oocyte development, germline cells dump their content into the growing oocyte and somatic support cells undergo apoptosis) strongly suggest that OSS cells accurately reflect the in vivo situation^31,32.

Somatic Piwi bound piRNAs are ~22-30nt long and ~75% carry a 5’ Uridine, a pattern found for several Argonaute family proteins^20,25,27. Over 60% of somatic piRNAs map the genome multiple times. This comes as no surprise given that 70% of them map to annotated transposons or transposon fragments, a strong enrichment compared to the average transposon content of ~10% in the assembled genome (Box 3). The remaining 30% of piRNAs map to non-annotated regions and protein coding genes. Given the observed piRNA composition, the cell evidently selects specific RNAs for piRNA processing. Significant insight into this and therefore piRNA biogenesis in general has been extracted from piRNAs that map to transposons and gene exons:

BOX3. Transposable elements in Drosophila.

The 180 Mb Drosophila genome harbors over 100 transposon families and members of all major classes (LINE and LTR type retro-elements as well as DNA-type elements) are represented⁹⁸,⁹⁹. Release 5 of the assembled genome contains 117 Mb of euchromatic and ~24 Mb of heterochromatic sequence. Assembly of the heterochromatic portion was an important prerequisite for identifying piRNA clusters¹⁰⁰,¹⁰¹. Annotated transposon sequences make up 7% of euchromatin and ~75% of heterochromatin. Most heterochromatic insertions, however, are sequence fragments and around two thirds of the 560 full length and thus presumably active elements are located in euchromatin. The genomes of other Drosophilid species contain similar transposon-families. Both, transposon load and their individual sequences are, however, typically species specific¹⁰². Consequently, piRNA clusters differ in their content and serve as species-specific repositories of transposons that are or have been active in a population³¹.

Transposon derived piRNAs

In the soma, over 90% of transposon annotated piRNAs are antisense to active transposon transcripts²¹, a clear contrast to the siRNA pathway, where sense and antisense small RNA populations are equally abundant^26,33–36. If mapped across transposon transcripts, piRNAs typically cover the entire sequence^31,32. No obvious patterns indicate preferential processing from certain regions, an indication that RNA structure does not trigger piRNA biogenesis. However, in some cases, piRNA profiles exhibit pronounced boundaries and certain transposon regions do not give rise to piRNAs. Insight into the genomic origin of piRNAs offered a coherent explanation for both, the antisense bias of piRNAs and the irregularities of piRNA profiles across certain elements^20,31.

About 15% of somatic transposon-derived piRNAs map uniquely to the genome and only these allowed the confident identification of genomic piRNA origins²⁰. This led to two conclusions: First, piRNAs appear to originate predominantly from broken transposon copies or their sequence fragments rather than from active copies. And secondly, piRNA generating transposon sequences are densely packed in a few genomic loci. These so-called piRNA clusters span dozens to hundreds of kilobases in length. They demarcate the regions with the highest density of broken, mutated and therefore immobile transposon fragments in the entire genome^20,37. piRNA clusters are a conserved hallmark of piRNA pathways, though their repeat content varies widely^38–41. In somatic ovarian cells, two piRNA clusters dominate and both are located on the X-chromosome roughly at the euchromatin/heterochromatin boundary. The larger one is the genetically identified flamenco locus, the smaller one is referred to as cluster 20A according to its cytology^{20,21,28–31}. From both clusters, piRNAs are derived only from one genomic strand, arguing for uni-directional transcription. Moreover, ~90% of the transposon fragments in flamenco and 100% of those in cluster 20A are oriented antisense to the transcription direction. This immediately explains the massive antisense bias of transposon-derived piRNAs.

The flamenco cluster appears to be only expressed/processed in somatic ovarian cells³¹. Interestingly, most transposon fragments in flamenco belong to the gypsy-family of retrotransposons, precisely those that invade the germline via the somatic niche^16–19,31.

These observations have led to a model in which the somatic piRNA pathway stores sequence information of transposons in specialized genomic regions. Their uni-directional transcripts are parsed into piRNAs, which -after loading into Piwi- allow the silencing of complementary transposons in trans^30,31. In this scenario, insertions of gypsy-type elements antisense to the direction of cluster transcription were positively selected during evolution. Strong support for this model stems from an analysis of ZAM fragments located within flamenco: Only sequence regions of ZAM that are found within flamenco give rise to abundant piRNAs³¹ (Fig.3E). piRNA production in the soma should therefore be independent of the expression of active elements. Indeed, levels of gypsy-derived somatic piRNAs are not influenced by the presence of active gypsy elements in the genome⁴².

Figure 3. piRNA profiles along clusters and transposons are tissue specific.

This figure illustrates the pronounced differences in piRNA pools found in somatic and germline cells of the Drosophila ovary. Basis for these differences are tissue specific transcription of piRNA clusters and presence of the ping-pong cycle in germline cells only. Somatic graphs are based on the OSS cell data from ref.43 and germline graphs are based on early embryo libraries from ref.66. To enable comparison of these populations, profiles were always normalized to 1 millionn sequenced repeat-derived 23-30nt small RNAs. Panels A-D indicate that flamenco is a soma-specific piRNA cluster while cluster 42AB is germline specific (cluster coordinates are shaded in light grey). Cluster 20A is processed into piRNAs in both cell-types. Also apparent is the unidirectional transcription of flamenco and cluster 20A while cluster 42AB is transcribed in both orientations. In each panel, only genome-unique piRNAs were used and a 200 nt sliding window with step size of 20 nt was applied. Sense and antisense piRNAs are displayed as upwards and downwards peaks, respectively (E-G) Shown are schematics of the LTR retrotransposon ZAM, Blood and Max-Element. Blue bars display the respective transposon fragments found in piRNA clusters (antisense ZAM fragments within flamenco, a complete antisense Blood element in cluster20A and Max fragments in cluster 42AB). Transposon cartoons and cluster fragments are length matched to the piRNA profiles shown below. Panels H-M show profiles of somatic and germline piRNAs mapping to ZAM, Blood and Max. For each graph, piRNAs mapping with up to three mismatches to the indicated element were pooled. ZAM is a proto-typical element expressed and silenced in somatic cells, while Max is apparently only silenced (and presumably transcribed) in germline cells. Blood silencing is active in both cell-types. The ZAM fragments present in the flamenco piRNA cluster (blue) are in striking agreement with the observed piRNA profiles, suggesting that they are the major source of ZAM piRNAs. Similarly, piRNA profiles for Blood and Max are consistent with their respective fragments in piRNA clusters 20A and 42AB. Ping-pong signatures are significant only for Blood and Max in the germline samples (not shown).

Besides flamenco and cluster 20A several other, often smaller piRNA clusters have been cataloged based on OSS piRNAs²¹. These are, however, not yet assembled into chromosomal contigs. We note that flamenco ends in a genomic gap of unknown size and that some of these cluster fragments therefore likely correspond to pieces of flamenco.

piRNAs from exons

Based on the suspicious localization of piRNA clusters at the euchromatin/heterochromatin boundary, one might suggest that certain chromatin marks allow flagging cluster transcripts for piRNA biogenesis. This model was challenged with the surprising discovery that transcripts from several hundred genes are substrates for piRNA biogenesis and are the source of nearly 10% of somatic piRNAs^22,43. About 95% of genic piRNAs are in sense orientation to the host transcript and typically originate from exons indicating that mature mRNAs are processing substrates.

Only a subset of cellular mRNAs gives rise to piRNAs and there is no apparent correlation between transcript abundance and piRNA levels⁴³. Since exonic sense piRNAs will typically have no fully complementary targets within the cell, their function is unknown. It has been suggested that some exonic piRNAs target cellular transcripts via incomplete pairing²². The predicted target sites, however, are located in the intron of the target gene. It remains to be shown, whether this allows significant target regulation and how tolerant target recognition is towards incomplete complementarity between small RNA and target. Alternatively, the cell modulates expression of the piRNA host gene directly, as some of the mRNA transcripts are consumed during piRNA biogenesis. Interestingly, the gene giving rise to most piRNAs encodes the transcription factor Traffic jam, which is required for Piwi expression in somatic support cells, indicative of a classic negative feedback loop^22,43.

Primary piRNA biogenesis

The linear biogenesis of piRNAs from precursor transcripts into PIWI proteins has been termed “primary piRNA biogenesis”²⁰. In somatic support cells of the gonad piRNAs seem to be exclusively generated via primary processing. Little is known about this process at the mechanistic level. The above-mentioned features of piRNAs strongly suggest that single stranded transcripts (originating from piRNA clusters and genes) are substrates for the processing machinery. A P-element insertion at the beginning of flamenco abrogates piRNA production over the entire 180kb cluster, strong evidence for a long, single stranded transcript^20,30,31. Moreover, piRNA biogenesis is independent of Dicer²⁶, the key enzyme in the miRNA and siRNA pathways where dsRNAs serve as trigger molecules. piRNA profiles across exons or clusters do not correlate with any obvious RNA secondary structure elements. Nevertheless, pronounced peaks of genome-unique piRNAs across clusters are apparent (Fig.3A,D). Differences in sequence content between analyzed piRNA clusters and reference genome and the fact that only some cluster regions can generate “genome-unique” piRNAs are the likely basis for these irregularities. It therefore appears as if precursor transcripts are randomly processed into piRNAs.

It is entirely unclear, how the cell selects cluster transcripts and those from a subset of genes for piRNA biogenesis. Are these transcripts marked in any special way for biogenesis? An experimental entry point into this question might well reside in the pool of genic piRNAs. In some cases the level of genic piRNAs per kilobase is approaching that of flamenco-derived piRNAs, indicating that genic piRNAs are not merely noisy by-products of cellular RNA metabolism. Somehow, these transcripts are special and it will be important to decipher the underlying molecular reason, be it sequence motifs or features like RNA half-life or translation efficiency.

The precise subcellular location for piRNA biogenesis is unknown. Piwi is enriched in the nucleus. Nevertheless, accumulating evidence suggests that primary piRNA biogenesis occurs in the cytoplasm. First, an overwhelming proportion of genic piRNAs originates from the 3’UTR with the first piRNAs mapping shortly downstream of the stop codon^22,43. This indicates that ribosomal association precedes piRNA processing. Secondly, an N-terminally truncated Piwi protein that cannot localize to the nucleus is loaded efficiently with piRNAs²².

Though variable in length, piRNAs with the same core sequence typically share the same 5’ end and differ in their 3’ ends. The first base shows a strong bias for Uridine. Preferences for 5’ nucleotides are common among Argonaute proteins^44–46 and a recent study lends structural support for the ability of Argonaute proteins to read out the identity of the bound RNA’s first base⁴⁷. In a random processing model, it seems likely that Piwi selectively stabilizes pre-piRNAs starting with a 5’ Uridine. This might well explain observed local irregularities in piRNA profiles. In a second step, 3’ trimming of the pre-piRNA would generate the heterogeneous 3’ end. The footprint of Piwi on the pre-piRNA would determine piRNA length, explaining why piRNA populations bound to different PIWI family proteins differ in their length²⁰. According to this, piRNA precursors are loaded as single stranded RNAs into Piwi. This is in contrast to siRNAs and miRNAs, which are loaded as small RNA duplexes into Argonaute proteins, after which one strand is removed^48–52. A recent study from S. pombe, however, indicates that the proposed piRNA biogenesis model is not that exotic in the end: Though fission yeast Argonaute is primarily loaded with Dicer products (small RNA duplexes), it appears that initially it is loaded with so-called primal RNAs⁵³. Strikingly, primal RNAs are preferentially derived from 3’UTR regions of cellular transcripts and centromeric repeats in a Dicer independent manner and appear to be trimmed at their 3’ end, potentially by the exosome⁵³. Only upon target interaction and cleavage, an RNA dependent RNA Polymerase (RdRP) converts the target transcript into dsRNA, providing the substrate for Dicer processing into the much more abundant siRNAs. The resemblance of primal RNAs to primary piRNAs is provoking, though no RdRP dependent amplification seems to participate in the piRNA pathway.

The proteins involved in primary piRNA biogenesis are unknown with the exception of Zucchini; a predicted nuclease with a phospho-lipase D domain^22,31,54. At which step Zucchini acts during biogenesis is, however, unclear. Though multiple other proteins have been identified as essential piRNA pathway members, genetic studies indicate that they are selectively involved in the more complex germline piRNA biology^31,32,55. Without a doubt, several as yet to be identified factors must participate in biogenesis, loading and function of primary piRNAs and the availability of the OSS cell line⁵⁶ promises rapid progress towards their identification and characterization.

The germline piRNA pathway and piRNA ping-pong

Considerable evidence indicates that the linear primary piRNA pathway feeding into Piwi is also active in germline cells^31,32. piRNA biology in the germline is, however, much more complex: Ovarian germline cells express besides Piwi also Aubergine and AGO3, two related PIWI family proteins^20,32,57,58. Sequence analysis of piRNAs selectively bound to Piwi, Aub and AGO3 revealed the existence of a sophisticated piRNA amplification loop that acts in parallel or on top of the above described primary piRNA pathway (Fig. 2). The central players in this so-called ping-pong cycle are Aub and AGO3, which localize to the cytoplasm of germline cells with an accumulation around the nucleus^20,58. In the prevailing model, Aub is guided via an antisense piRNA to a sense transcript from an active transposon. Subsequent slicer cleavage of the target transcript triggers production of a novel sense piRNA, which is loaded into AGO3. The AGO3-piRNA complex in turn cleaves complementary piRNA cluster transcripts. This prompts biogenesis of a novel antisense and Aub-bound piRNA, whose sequence is identical to the initiator piRNA. As Aub and AGO3 presumably act catalytically⁵⁸, the ping-pong cycle amplifies silencing competent piRNAs with the loop acting efficiently only in the presence of a target transcript (active transposon message). Indeed, ping-pong piRNAs are the most abundant cellular piRNAs²⁰. A key conceptual difference to the primary piRNA pathway is that piRNA biogenesis in the germline depends in part on target expression. Elegant genetic experiments have hinted at this even before piRNAs were discovered^59,60. Signatures of the ping-pong cycle have been found in sponges, planaria, moths, fish, frogs and mammals^40,61–65 (Box 4). It is therefore one of the hallmarks of the piRNA pathway. Target dependent amplification of a small RNA response is common among small RNA pathways in fungi, plants and nematodes⁹. Here, however, cleavage of the target triggers dsRNA synthesis by RdRP enzymes, generating Dicer substrates. Most animals lack RdRP enzymes and the piRNA pathway utilizes instead sense and antisense RNAs from different transcripts and couples them via reciprocal Slicer cleavage.

Figure 2. In germline cells, the primary piRNA pathway and the Ping-Pong amplification loop are active.

Shown are representative examples of germline piRNA sources and targets originating from the 2^nd chromosome (drawn at the top). Colored boxes show primary piRNA biogenesis from cluster 42AB (yellow), the adaptive module of the target dependent ping-pong amplification loop (red), expression and silencing of a typical active LTR-retrotransposon (Max-Element; blue) and the contribution of maternally inherited piRNAs (green). (yellow box) Cluster 42AB contains transposon fragments in both orientations and is bi-directionally transcribed. During primary piRNA biogenesis, cluster transcripts (red) presumably generate sense and antisense piRNAs. Unknown mechanisms parse the long piRNA precursor transcripts into shorter fragments that are loaded onto PIWI family proteins (Piwi, probably Aub and potentially AGO3). Piwi and Aub probably select RNA fragments with a 5’ Uridine (1U). Subsequently, the 3’ tail of pre-piRNAs are trimmed and 2’OH-methylated to generate mature piRNAs. piRISCs with antisense piRNAs are competent to silence sequence complementary transcripts of active transposons. Primary piRNA biogenesis in the germline is likely similar to the one in somatic cells. The blue box shows an active copy of the Max-Element (LTR-retrotransposon) and its transcribed sense transcript that is silenced by complementary piRISCs. The red box summarizes the ping-pong cycle. An Aub complexed piRNA (red) that is antisense to an active sense Max transcript (dark blue) guides slicing (scissors) of the transposon RNA, precisely 10nt downstream of its 5’ Uridine. The sliced Max transcript is predicted to be loaded onto AGO3 and typically has a profound bias for an Adenine at position 10 (10A). The AGO3 bound pre-piRNA is 3’ trimmed and 2’-OH methylated. This mature AGO3-piRNA complex in turn cleaves complementary cluster transcripts and triggers production of an Aub-loaded antisense piRNA, whose sequence is identical to the initiating piRNA. It is currently impossible to experimentally distinguish between Aub-piRNA complexes generated via primary piRNA biogenesis or via ping-pong. Weak ping-pong signatures exist between AGO3 and Piwi and could indicate that Piwi (besides primary biogenesis) also receives piRNAs via AGO3 mediated target slicing. (green box) At the end of oogenesis, mature Aub- and Piwi-piRNA complexes (to a lesser extent also AGO3) are efficiently loaded into the oocyte. Maternal Aub and to a lower extent also Piwi localize to the posterior pole of the mature egg, where the future germline will form. Maternal piRNAs might serve important roles in the starting phase of the pig-pong cycle.

BOX4. Commonalities and differences among animal piRNA pathways.

Based on primary sequence analyses, PIWI proteins are found throughout the animal kingdom. They are typically expressed in germline cells but in lower invertebrates also in cells responsible for regeneration (e.g. neoblasts in Planaria⁶²,¹⁰³). Primary piRNA biogenesis and signatures of the ping-pong amplification cycle are found in species ranging from sponges to mammals, indicating an ancient origin of the pathway’s key features⁶¹. A notable exception is C. elegans, where the two PIWI family proteins PRG1 and PRG2 are expressed in gonads yet bind 21U RNAs, a different class of small RNAs⁴⁵,⁹⁵–⁹⁷. These appear to have distinct biogenesis features, do not exhibit ping-pong signatures and have a largely unclear function and target spectrum.

In mammals, three PIWI family proteins are expressed in testes and only one in ovaries¹⁰⁴–¹⁰⁷. Interestingly, the pathway seems largely dispensable for oogenesis, where an endo-siRNA pathway centered on AGO proteins cooperates in silencing transposable elements⁹³,⁹⁴. During mouse spermatogenesis, the three PIWI family proteins MIWI2, MILI and MIWI are expressed in different, yet overlapping temporal domains. MIWI2 and MILI are the key players in the genome defense pathway and their bound piRNAs show signatures of ping-pong and primary piRNA biogenesis¹⁰⁸,¹⁰⁹. MIWI2, the only nuclear mouse PIWI family protein is suggested to guide de novo DNA methylation at transposon loci, a process so far only reported for the mammalian piRNA pathway¹⁰⁸,¹⁰⁹. MIWI is expressed only after the pachytene stage of meiosis and binds primary piRNAs derived from a distinct set of piRNA clusters. Pachytene piRNAs are not enriched in transposon sequences, accumulate to very high levels and have an unclear function, but presumably distinct from transposon silencing³⁸,³⁹,⁴¹.

Germline piRNA clusters

RNAs in early embryos –prior to zygotic transcription- reflect by and large the pool of germline transcripts made during oogenesis^31,66. piRNAs from young embryos are thus the best proxy for the germline specific piRNA pool. Germline piRNAs originate predominantly from several piRNA clusters but also from transcripts of active elements, in accordance with the ping-pong model. Evidence for this is best documented for the I-element (LINE family), where cluster resident fragments and active elements have sufficiently diverged at the sequence level to allow distinguishing them⁶⁶.

Just like soma clusters, germline clusters are strongly enriched in transposon fragments and the most prominent ones map to euchromatin/heterochromatin boundaries²⁰. As a much broader spectrum of transposable elements (LINE-, LTR- and DNA-type elements) is highly active in the germline, it comes as no surprise that germline piRNA clusters contain a more diverse collection of transposon fragments³¹. Figure 3 depicts the soma specific flamenco cluster, the germline specific cluster 42AB and the shared cluster 20A. Cluster 20A is the only germline piRNA cluster that resembles flamenco as it is uni-directionally transcribed and contains only antisense transposon fragments^20,31. In the germline, it loads preferentially Piwi and to a lesser extent also Aub. All other germline clusters spawn piRNAs from both strands indicating bi-directional transcription and alleviating the pressure for transposons to integrate in a biased orientation as observed for flamenco³¹. piRNAs originating from these clusters are loaded into all three PIWI family proteins but absolute numbers cannot be derived as only a minority of piRNAs can unambiguously be mapped to clusters. Finally, germline piRNA clusters are also found at telomeres, where abundant piRNAs are derived from both, the telomeric arrays of HeT-A, TART and TAHRE elements as well as from subtelomeric satellite repeats²⁰.

It is unclear whether piRNA cluster transcripts are essential for ping-pong or whether any sense and antisense transcripts could engage in it. Similarly unknown is how the cell prevents auto-amplification of piRNAs derived from bi-directionally transcribed clusters. Best evidence for this stems from the analysis of I-element piRNAs: I-element fragments within cluster 42AB give rise to high piRNA levels only in strains with active elements⁶⁶. Perhaps, cluster transcripts are physically isolated in specific cellular domains and presented only to selected protein complexes to guarantee accurate progressing through the biogenesis cycle.

The connections between primary piRNA biogenesis and the ping-pong cycle are only poorly understood. The ping-pong signature is mostly confined to Aub/AGO3 and Aub/Aub pairs^20,32. Ping-pong is almost entirely lost in aub mutants³¹. While Aub/Aub ping-pong prevails in AGO3 mutants, the resulting piRNA levels are severely diminished³². Piwi on the other hand is dispensable for the ping-pong cycle³¹. Nevertheless, a weak but significant ping-pong interaction occurs between Piwi and AGO3^20,32, suggesting that Piwi –while not providing input- could be a recipient in the cycle (Fig. 2). This might explain the antisense bias of Piwi bound piRNAs originating from germline clusters²⁰ and might also explain the loss of nuclear Piwi in late stage ovarioles lacking AGO3³². In such a model, primary piRNA biogenesis must also load Aub or AGO3. In fact, in Drosophila testes, germline cells express only Aub and AGO3, while Piwi is expressed in somatic support cells only⁶⁷. It therefore remains to be shown, how primary piRNA biogenesis feeds into the ping-pong cycle.

The maternal piRNA pool

During oogenesis, the oocyte is connected to the 15 nurse cells via cytoplasmic bridges. Ultimately, nurse cells dump their cytoplasm into the maturing oocyte (Fig. 1A). Piwi and Aub localize to the oocyte’s posterior pole, where the primordial germ cells of the embryo will form^57,66,68. Maternal piRNAs thus form a protective layer against transposons even before transcription initiates in future germ cells. Genetic experiments have suggested the existence of a maternal factor with essential roles in transposon silencing^3,4. In this so-called hybrid dysgenesis phenomenon, crosses between naïve females and males carrying a novel transposon produce sterile offspring, while the reciprocal cross does not. Small RNA sequencing approaches have shown that inheritance of maternal piRNAs is required for an efficient ping-pong response in the F1 generation⁶⁶. Three possibilities could explain this observation: (1) Maternal piRNAs are required to kick-start the ping-pong cycle. (2) Maternal piRNAs have an essential influence on the chromatin status of piRNA clusters and/or transposons. (3) The cellular Aub and AGO3 pools are limiting and in the absence of maternal piRNAs for a certain element, the low level of primary piRNAs is unable to compete with piRNAs abundantly inherited maternally.

In summary, the germline piRNA pathway is considerably more elaborate then the linear somatic piRNA pathway. Also here, we largely lack insight into the molecular and cellular details. Genetic studies have identified multiple proteins involved in the piRNA pathway^{23,31,32,54,55,57,69–78} and several are specifically required for the germline piRNA pathway^31,32,55. Amongst those are several RNA helicases but also proteins with unknown functions. Recent studies have linked the piRNA pathway to Tudor biology⁷⁹. Tudor domains bind symmetrically methylated Arginines in Aub, AGO3 and potentially Piwi^80–82. The Drosophila genome encodes at least twenty Tudor-domain containing proteins and many of these are selectively expressed in the germline (unpublished observations). Given this and considering the complexity of piRNA biology, we expect the number of proteins with essential roles in this genome defense system to increase considerably.

Outlook

Research in the piRNA field is in the paradoxical situation that we understand a lot about conceptual frameworks but that we lack almost every mechanistic and molecular insight. A great deal of understanding other small RNA pathways has emerged from in vitro assays. No such attempt has been reported for the piRNA pathway, probably reflecting the complexity of this approach. Without a doubt though, in vitro systems coupled with genetics and structural approaches will be essential to understand the order and dynamics of the molecular events during piRNA biogenesis and silencing.

The second challenge will be to understand the nature of piRNA cluster transcripts and to decipher the protein-RNA network that forms the basis of the piRNA pathway. Here, we expect rapid progress by combinations of next generation sequencing approaches with technological advances in determining RNA-protein interactions^83,84. All in all, these are exciting days for everybody working on this fascinating genome defense systems and if recent years are a measure, many surprises are yet to come.