Expression of Piwi family members during germline development
The mouse genome contains three Piwi family members: Mili
is expressed from the pachytene stage of meiosis to the haploid round spermatid stage (Deng and Lin, 2002
) (). MILI is also present during meiosis, at which point both MILI and MIWI interact with an extremely abundant class of small RNAs, the pachytene piRNAs (Aravin et al., 2006
; Girard et al., 2006
). These are derived from specific genomic loci and form a complex population of small RNAs that match only to those sites from which they are derived. The function of this sub-class of piRNAs is elusive.
Expression of Piwi proteins though germ cell development
is also expressed earlier in development. Just before entry into meiosis and the onset of pachytene piRNA expression, MILI binds piRNAs, which are different in character and genomic origin from meiotically expressed piRNAs (Aravin et al., 2007b
). These are derived from a set of clustered loci that are repeat-enriched and thus give rise to small RNA populations corresponding to transposons.
We investigated Mili and Miwi2 expression both by using specific antibodies raised against each family member and using transgenic animals that express GFP- and myc-tagged MILI and MIWI2 proteins under the control of their endogenous promoters. For each protein, examination of several GFP and myc-transgenic lines and comparison to results obtained with antibodies against native proteins revealed essentially identical patterns.
We analyzed the expression of Mili
during germ line development in male and female mice. Mili
expression could be detected in both sexes as early as 12.5 dpc, a time when migrating primordial germ cells (PGCs) have reached the somatic genital ridge (, not shown). MILI localized to numerous perinuclear cytoplasmic granules in male PGCs (). These structures were remarkably similar to nuage, which contain the Piwi proteins, AUB and AGO3, in Drosophila
female germ cells. MILI was also detected in female PGCs and localized to cytoplasmic granules (). Mili
expression continued in both male and female germ cells after birth (Fig. S1A
). In ovary it localized to cytoplasmic granules of both arrested and growing oocytes. In adult testis, it was expressed throughout spermatogenesis until the round spermatid stage at which point the majority of MILI localized in a single prominent granule, the chromatoid body, as evident from co-localization with chromatoid body marker MVH (DDX4) (Fig. S1B
MIWI2 could be detected in male germ cells beginning around 14.5–15.5 dpc but was absent from female germ cells (). MIWI2 was present in the nucleus as well as the cytoplasm. In the cytoplasm, MIWI2 occupied granules similar to but fewer in number than those containing MILI. From 15.5 dpc and until birth both proteins were present in male germ cells (). We did not detect MILI or MIWI2 in somatic cells of the embryonic gonad. MIWI2 expression declined soon after birth reaching undetectable levels in 4-day old mice. Remarkably, the very short window of MIWI2 expression during male germ cell development (15.5 dpc-3 dpp) corresponds to the time of cell cycle arrest and de novo DNA methylation.
MILI and MIWI2-bound piRNA populations during germline development
MILI and MIWI2 complexes were recovered from male embryonic gonads isolated from 16.5 dpc embryos. Each protein associated with piRNAs of a specific size. MILI bound ~26 nt. RNAs (Aravin et al., 2006
; Aravin et al., 2007b
), and MIWI2 associated with ~28 nt RNAs ().
We prepared libraries of piRNAs from MILI and MIWI2 complexes at 16.5 dpc and 24–33 nt total RNAs from the same stage. Additionally, we cloned libraries from 24–33 nt RNAs and MILI complexes at days 2 and 10 after birth. The small RNA libraries were sequenced, typically yielding 2–3 million small RNA reads per library. Between 40 and 70% of sequences matched perfectly to the mouse genome, and these were considered for further analysis (Suppl. Table 1
). MILI and MIWI2-associated sequences showed a normal distribution with peaks at 26 and 28 nt respectively (). The profile of total cellular small RNA suggested that MILI-bound piRNAs are slightly more abundant than MIWI2 piRNAs.
At 16.5 dpc 47.5% of all cellular small RNAs were derived from transposon sequences ( and Table S1
). The other small RNAs represent fragments of abundant, larger non-coding RNAs (29.6%), sequences derived from un-annotated genomic regions (20.3%) or exons of protein encoding genes (2.5%). The overall fraction of transposon-derived piRNAs remained relatively stable during development from 16.5 dpc to 10 dpp. However, different types of transposons showed distinct patterns. The fraction of LTR and LINE-derived piRNAs decreased while SINE-derived sequences increased during development. The fraction of exon-derived small RNAs also increased substantially (from 2.5% at 16.5 dpc to 18.6% at 10 dpp). In Mili
knock-out animals, LTR and LINE-derived small RNAs were almost completely eliminated indicating that they represent bona fide
piRNAs. SINE and exon-derived small RNA were decreased in abundance but not eliminated in knock-outs suggesting that small portion of this class derives from degradation product of SINE-containing and genic transcripts.
Both MILI and MIWI2 associated with repeat-derived piRNAs during prenatal development (). However, MIWI2 demonstrated greater specificity for transposons. 76% of MIWI2-bound piRNAs mapped to LTR and LINE retrotransposons as compared to 45.7% for MILI-bound species. MILI complexes also contained piRNAs derived from exons of protein-coding genes (4.8%). As observed for total RNA profiles, the fraction of MILI-bound LTR and LINE piRNAs decreased and SINE and exon-derived piRNAs increased from 16.5 dpc to 10 dpp. A large fraction of total cellular small RNAs (48.7%) and MILI-bound piRNAs (60.1%) could be mapped to a unique genomic position (Fig. S2
). Both total small RNAs and MILI-associated piRNAs contained a subset (14–17%) that matched highly repetitive sequences (100 or more genomic mappings). These dramatically decreased in Mili
mutants. In MIWI2 complexes the fraction of uniquely mapped piRNAs was lower (35.6%) and the highly repetitive fraction was substantially higher (30.5%).
Overall, our analysis revealed that both MILI and MIW2 bind transposon-derived piRNAs during embryonic development and that the fraction of piRNAs that match active transposons classes decreases after birth. Interestingly, the partners of the two Piwi proteins are different with MIWI2 complexes being particularly enriched in transposon-derived piRNAs.
A substantial fraction of piRNAs matched the three major classes of transposable elements present in mammalian genomes: LTR, LINE and SINE retrotransposons. We analyzed piRNAs derived from the representative element of each class that produced the largest number of small RNAs: the IAP LTR-retrotransposon, LINE1 and SINE B1. During development, the abundance of LINE1 piRNAs decreased and SINE increased, while IAP piRNAs showed a double-peaked with higher levels before birth and at 10 dpp (). LINE- and IAP-derived piRNAs were associated with both MILI and MIWI2 (). Interestingly, for both elements piRNAs were almost equally distributed among MIWI2 and MILI complexes (). Sequences derived from exons were almost exclusively bound to MILI ().
To probe the functions of Piwi-bound small RNAs, we analyzed strand orientation of transposon- and gene-derived piRNAs. Exon-derived piRNAs were highly enriched for sense sequences. All three transposons produced substantial numbers of antisense piRNAs, with different elements showing different characteristics (). For LINE1, piRNAs were generally enriched in antisense sequences throughout development. In prenatal testis MILI and MIWI2 had opposite strand preferences with MILI binding more sense and MIWI2 binding more antisense piRNAs. Interestingly, strand orientation of LINE1 piRNA in MILI complexes was reversed after birth. IAP and SINE piRNAs were nearly equally divided between sense and antisense sequences in both complexes and in total piRNA populations in prenatal cells. However at 10 dpp, sense IAP and SINE piRNAs became predominant. For all three transposons the fraction of antisense sequences dramatically decreased in MILI-deficient animals.
These results indicate that LINE1 piRNAs are sorted into MILI and MIWI2 complexes according to strand orientation, similar to what is observed for Piwi family members in Drosophila
(Brennecke et al., 2007
). However, this bias was slight for SINE B1 and was absent for IAP. These data also reveal that strand bias even within a given Piwi complex can be dynamic during development.
We next analyzed the distribution of piRNAs along transposon consensus sequences (). Notably, MILI-bound LINE1 piRNAs were enriched in antisense and biased toward the 5’ end of LINE1 in prenatal germ cells. At 10 dpp, piRNAs were less strand biased and mapped more frequently toward the 3’ ends of the elements, likely reflecting the higher abundance of those sequences in the genome (Kazazian, 2004
). The enrichment for piRNAs mapping to LINE1 5’ ends at 16.5 dpc implies that at this developmental stage piRNAs are processed from full-length, potentially active copies.
The ping-pong cycle in prenatal piRNAs
Two different mechanisms, primary processing and ping-pong amplification, have been proposed to generate piRNAs (Brennecke et al., 2007
), reviewed in (Aravin et al., 2007a
). Primary processing samples single-stranded piRNA precursor transcripts generating a diverse set of piRNA sequences that share a preference for 5’ uridine (1U). Pachytene piRNAs are exclusively produced by the primary processing mechanism. A subset of pre-pachytene piRNAs in mouse (Aravin et al., 2007b
) and a substantial fraction of Drosophila
piRNAs (Brennecke et al., 2007
) are generated by a mechanism that depends upon the endonuclease activity of Piwi proteins and that is referred to as the ping-pong amplification cycle.
The ping-pong cycle requires the presence of transcripts that are complementary to primary piRNAs. Recognition by primary piRNAs guides the endonuclease activity of Piwi proteins, which cleave the transcript 10 nucleotides from the 5’ end of the original piRNA (Brennecke et al., 2007
; Gunawardane et al., 2007
). This event generates 5’ end of a new secondary piRNA. These show a strong bias for adenine at position 10 (10A) complementing the 1U bias of primary piRNAs (). Secondary piRNAs can also generate new piRNAs by recognizing and cleaving complementary transcripts to regenerate the piRNA that initiated the cycle. Thus, only secondary piRNAs are enriched for 10A. Although piRNAs with the sequence identical to original primary piRNA can be created during the cycle (following cleavage by a 10A secondary piRNA), these 1U-biased species remain reflective of a primary species that initiated the cycle. Thus, we consider the 1U population to be reflective of primary piRNA biogenesis.
Ping-pong amplification in prenatal piRNAs
We investigated the existence of the ping-pong cycle in prenatal germ cells and the roles of MILI and MIWI2 in this process. Tracking a specific feature of ping-pong piRNA pairs, the 10nt offset between 5’ ends of piRNAs, showed that both MILI and MIWI2 participate in the amplification cycle. The most prevalent signatures indicated MIWI2-MIWI2 and MIWI2-MILI cycles ().
, Piwi proteins participating prominently in the ping-pong cycle show piRNA strand specificity (Brennecke et al., 2007
). We tested whether similar characteristics defined the ping-pong cycle in prenatal testis. Ping-pong pairs where the sense strand associated with MILI and antisense with MIWI2 were more abundant than pairs with the opposite character (). Surprisingly, this held true not only for LINE1 piRNAs that are generally asymmetrically distributed in MILI and MIWI2 complexes, but also for IAP piRNAs, that do not show a protein-dependent strand bias overall.
As was seen for strand orientation, MILI and MIWI2 complexes also discriminated primary and secondary piRNAs. We calculated the preference (Primary/Secondary, P/S, ratio) by taking the number of piRNAs that show 1U but no 10A bias (primary-like) and dividing it by the number that show a 10A but no 1U bias (secondary piRNAs). It should be noted that this approach ignores all sequences with both 1U and 10A, as these cannot be assigned to primary or secondary categories. Prenatal piRNAs are strongly enriched in secondary sequences as compared to pachytene piRNAs, which appear to be generated exclusively by primary processing (P/S ratios of 5.13 and 21.2 respectively). LINE1 and IAP piRNAs showed strong signals for secondary sequences as compared to exon-derived piRNAs, which given a lack of antisense information must be generated by primary processing (). MIWI2 complexes were ~2 fold enriched in secondary piRNAs as compared to MILI. Overall our data indicate distinct roles for MILI and MIWI2 in the piRNA-processing pathway. MILI is biased toward primary piRNAs and 1U-containing piRNAs generated in the ping-pong cycle. MIWI2 is particularly enriched in secondary sequences.
Finally, we probed the correlation between the strand orientation of transposon piRNAs and their processing category (primary or secondary). In both MILI and MIWI2 complexes, antisense piRNAs were enriched for secondary sequences. MILI-associated sense piRNAs had a P/S ratio of 4.46 versus 2.02 for antisense, and MIWI2-associated sense and antisense piRNAs had P/S ratios of 2.54 and 1.40, respectively ().
Overall, these data are consistent with a model in which sense transcripts, most likely mRNAs of active transposons, represent the major substrate for primary processing and result in piRNAs associated with MILI. MILI-associated primary sense piRNAs recognize and cleave transcripts that contain transposon sequences in the antisense orientation and generate secondary piRNAs that join MIWI2 complexes. This is precisely opposite to the bias observed in Drosophila, where primary piRNAs are mostly derived from piRNA clusters and are enriched for antisense strands while secondary piRNAs are sense.
Ping-pong relationships between MILI and MIWI2 suggest a need for physical proximity. Co-localization of MILI and MIWI2 was investigated by immunofluorescence in 17.5 dpc germ cells. Though MIWI2 was mainly present in the nucleus, MIWI2-containing cytoplasmic granules co-localized with or were in close proximity to MILI-containing granules (). MILI granules were more abundant as compared to MIWI2 granules, and many MILI granules did not interact with MIWI2 foci. In Miwi2-deficient cells, MILI localization in cytoplasmic granules did not change (). In contrast, in Mili mutants MIWI2 re-localized from the nucleus to the cytoplasm where it was uniformly distributed rather than concentrating in granules (). Notably, we detected no MIWi2-associated piRNAs in Mili mutants () indicating that MIWI2 remains unloaded when MILI is absent. This epistatic relationship between MILI and MIWI2 supports the proposed directionality of their interaction wherein MILI initiates the cycle with primary piRNAs, and the production of secondary piRNAs associated with MIWI2 depends upon the prior existence of these species.
Interaction between MILI and MIWI2 complexes in germ cells
Genomic origins of prenatal piRNAs
Derivation from clustered loci in the genome has been a defining feature of piRNAs, and these loci play an important role in piRNA generation in both Drosophila
and vertebrates (Aravin et al., 2006
; Aravin et al., 2007b
; Brennecke et al., 2007
; Girard et al., 2006
; Lau et al., 2006
). To investigate the genomic origin of prenatal piRNAs we searched for unambiguously mapping sequences in close proximity in the genome. Using a threshold value of 10 piRNAs per kilobase, we identified 3399 clusters, which were ranked by their relative contributions to piRNA populations. Though the most prominent cluster gave rise to almost 10% of all uniquely mapped piRNAs, the individual contribution of each subsequent cluster dropped dramatically (). Combining all clusters yielded about 8 Mb of genomic space, which could accommodate only ~50% of all uniquely mapping piRNAs. This suggests fundamental differences from the pachytene piRNAs (Aravin et al., 2006
; Girard et al., 2006
) and even those pre-pachytene piRNAs that occupy Mili
at 10 dpp (Aravin et al., 2007b
). In fact, most prenatal clusters actually represent individual transposons and transposon fragments. Expression from all clusters was eliminated in Mili
Genomic origins of prenatal piRNAs
Analyzing the expression of clusters that gave rise to the greatest numbers of piRNAs indicated developmentally regulated expression patterns. Only cluster #2 on chromosome 10 persisted in its expression through day 10 dpp. Interestingly this cluster was also highly expressed in ovary ().
Unlike piRNA clusters identified in mammals thus far, 6 out of the 8 most prominent prenatal clusters generate piRNAs from both genomic strands (double-strand clusters). This arrangement is very similar to piRNA clusters in Drosophila
ovary (Brennecke et al., 2007
). The two most prominent clusters (#1 chr.7 and #2 chr.10) show profound strand asymmetry with the vast majority of piRNAs being derived from one genomic strand (single-strand clusters). As with Drosophila flamenco
, these clusters also showed enrichment for similarly oriented transposon fragments and for the generation of piRNAs that are antisense to transposon mRNAs (). Surprisingly, double-stranded clusters are not particularly enriched in transposon sequences. Both single-strand and double-strand clusters produced piRNAs in MILI and MIWI2 complexes (). However, even in double-strand clusters MILI-bound piRNAs were strongly biased for one of the two genomic strands (). Size profiles of total cellular piRNAs indicated that single-strand clusters produced significantly more MILI-bound piRNAs (), while double-strand clusters contributed to MILI and MIWI2 complexes equally ().
piRNAs derived from clusters participate in the ping-pong cycle (data not shown). For cluster-derived piRNAs, MIWI2 remained biased towards secondary piRNAs as compared to MILI (). Single-strand clusters produced more primary piRNAs in both MILI and MIWI2 complexes than did double-strand clusters. Indeed, the P/S ratio of MILI-bound piRNAs derived from the single-strand cluster on chr. 7 (16.21) was close to that of pachytene piRNAs (21.2).
If ping-pong amplification operates by interaction of piRNAs from single-strand clusters with transposon mRNAs in trans, MIWI2, which is enriched in secondary piRNAs, should contain more transposon-derived piRNAs than MILI. Indeed, for the two single-strand transposon-rich clusters (#1, chr. 7 and #2, chr.10), MIWI2 was particularly enriched in transposon-derived piRNAs (). The difference between MILI and MIWI2 complexes was even more dramatic when strand orientation was taken into account. For both clusters, MIWI2 was enriched in piRNAs that match transposons in the antisense orientation (). These data indicate that the ping-pong cycle strongly shapes piRNA populations and enhances the production of piRNAs that match transposons in the antisense orientation.
DNA methylation and Piwi/piRNA pathway
DNA methylation is critical to stable, epigenetically inherited silencing of transposons, and this is lost upon mutation of either MILI or MIWI2 (Aravin et al., 2007b
; Carmell et al., 2007
; Kuramochi-Miyagawa et al., 2008
). DNMT3L, a catalytically defective member of the DNA methyltransferase family, is essential for both proper transposon methylation and transposon repression (Bourc'his and Bestor, 2004
). DNMT3L acts together with the catalytically active de novo
methyltransferases, DNMT3A and DNMT3B, to establish methylation patterns (Chen et al., 2005
; Gowher et al., 2005
; Suetake et al., 2004
). We sought to order MILI and MIWI2 with respect to other pathway components. Since DNMT3L acts as a coordinator of methylation activities in the male germ line, we examined the integrity of the piRNA pathway in Dnmt3L
We immunoprecipitated MILI from testes of 10 dpp Dnmt3L-deficient animals and their wild-type littermates and found that piRNAs were still present in mutant animals (). Cloning and analysis of small RNA libraries showed that the fraction of LTR and LINE retrotransposon piRNAs increased, while the fraction of SINE piRNAs decreased in Dnmt3L mutants (). This was consistent with Northern blotting for an abundant IAP-derived piRNA, which also increased in abundance in the mutant ().
Links between the DNA methylation and piRNA pathways
Several lines of evidence indicated that increases in LTR and LINE piRNAs might be linked to de-repression of these elements and the increased capacity of transposon mRNAs for entry into the piRNA pathway. First, LINE and LTR elements lost methylation and silencing, showed increased expression and contributed a greater number of sense piRNAs in Dnmt3L mutants. In contrast SINEs (B1) were neither affected at the level of DNA methylation and expression nor contributed increasingly to piRNA populations in mutant animals (unpublished data). Moreover, LINE and LTR piRNAs that increased in Dnmt3L mutants corresponded to elements that are close to consensus and thus potentially expressed under circumstances where methylation was lost ().
As an example, piRNAs derived from IAP elements showed the greatest change in Dnmt3L mutants, and this was mainly due to a dramatic increase in sense small RNAs (). This pattern was also obvious when the distribution of piRNAs along the IAP consensus was displayed (). Normally, the ratio of primary to secondary sequences in MILI complexes decreases after birth and approaches that of MIWI2 in prenatal germ cells (, compare to ). In Dnmt3L-deficient animals the ratio of primary to secondary IAP piRNA increased ~ 6-fold as compared to wild-type. This strongly supports the model that mRNAs from IAP elements, which show increased expression in Dnmt3L mutants, flow into the piRNA pathway as a source of primary piRNAs.
Considered, these data support a model in which the piRNA pathway acts upstream of DNMT3L, and consequently DNMT3A and 3B, to help establish patterns of DNA methylation on repeat elements. This interpretation rests on the observation that while mutations in methyltransferase family members impact cytosine methylation, they leave the piRNA pathway largely intact, affecting only the composition of small RNA populations in manner that can be rationalized by the impact of loss of methylation on transposon expression. In contrast, loss of the piRNA pathway prevents the recognition and silencing of potentially active transposons by the DNMT3L pathway.