Adaptive Regulation of Testis Gene Expression and Control of Male Fertility by the Drosophila Harpin RNA Pathway

SUMMARY

Although endogenous siRNAs (endo-siRNAs) have been described in many species, still little is known about their endogenous utility. Here, we show that Drosophila hairpin RNAs (hpRNAs) generate an endo-siRNA class with predominant expression in testes. Although hpRNAs are universally recently evolved, we identify highly complementary protein-coding targets for all hpRNAs. Importantly, we find broad evidence for evolutionary divergences that preferentially maintain compensatory pairing between hpRNAs and targets, serving as first evidence for adaptive selection for siRNA-mediated target regulation in metazoans. We demonstrate organismal impact of hpRNA activity, since knockout of hpRNA1 derepresses its target ATP synthase-β in testes and compromises spermatogenesis and male fertility. Moreover, we reveal surprising male-specific impact of RNAi factors on germ cell development and fertility, consistent with testis-directed function of the hpRNA pathway. Finally, the collected hpRNA loci chronicle an evolutionary timeline that reflects their origins from prospective target genes, mirroring a strategy described for plant miRNAs.

INTRODUCTION

The conserved RNAi and miRNA pathways process double-stranded RNA (dsRNA) substrates into ~21 nt siRNAs and ~21–24 nt miRNAs, respectively. The details regarding the biogenesis and function of siRNAs and miRNAs differ across many well-studied organisms, but key features include the involvement of one or more RNase III enzymes (e.g., Dicer) to digest dsRNA precursors into small RNAs, which load into Argonaute effector proteins and guide them to targets (Yang and Lai, 2011).

RNAi may have emerged to restrict transposons and viruses and been subsequently repurposed via miRNAs to regulate endogenous genes. Generally speaking, miRNAs are defined as specifically processed small RNAs derived from endogenous foldback transcripts. Despite overall similarity of plant and animal miRNA pathways, they exhibit fundamental differences in biogenesis and function (Axtell et al., 2011). Indeed, miRNA biogenesis strategies have proven highly plastic and capable of interfacing with multiple ribonucleases to generate functional miRNAs (Lee et al., 2010; Maurin et al., 2012; Yang and Lai, 2011). Thus, it is conceivable that miRNA pathways emerged more than once during evolution.

Plant miRNAs often exhibit extensive complementary to target genes (Rhoades et al., 2002), leading to the notion that they often derive from their prospective targets (Allen et al., 2004). In support of this, evolutionary “young” plant miRNAs can exhibit extended target homology beyond their mature small RNAs and have heterogeneous hairpin lengths that can reach hundreds of nucleotides (Fahlgren et al., 2007; Rajagopalan et al., 2006). In contrast, animal miRNA hairpins are more homogenous, typically ~65–80 nt. Rare animal miRNAs exhibit perfect complementarity to targets, such as miR-127/miR-136:Rtl1 (Seitz et al., 2003) and miR-196:HoxD8 (Mansfield et al., 2004; Yekta et al., 2004). However, the predominant targeting strategy for animal miRNAs involves Watson-Crick pairing to nucleotides 2–8 of the miRNA, also known as the seed (Brennecke et al., 2005; Lai, 2002; Lewis et al., 2005). The preference for shorter hairpins and minimal target pairing in animals suggested their miRNAs might typically arise de novo and mostly lack functional targets upon birth (Bartel and Chen, 2004; Chen and Rajewsky, 2007).

Drosophila harbors diverse endogenous siRNAs, processed from several classes of progenitor transcripts (Okamura and Lai, 2008). These include from transposable elements (TEs), cis-natural antisense transcripts (cis-NATs), and long inverted repeat transcripts known as hairpin RNAs (hpRNAs) (Chung et al., 2008; Czech et al., 2008; Ghildiyal et al., 2008; Kawamura et al., 2008; Okamura et al., 2008a, 2008b). hpRNA stems are heterogenous and range from hundreds down to ~70 bp, meaning that the shortest hpRNAs resemble the longest canonical miRNAs in Drosophila (Okamura et al., 2008b; Ruby et al., 2007). Although hpRNAs bear extensive duplex imperfections, they are not substrates of the Dcr-1/AGO1 miRNA pathway and instead specifically transit the Dcr-2/AGO2 siRNA pathway. The Dicer cofactor Loquacious (Loqs) is involved in biogenesis of both miRNAs and siRNAs (including from hpRNAs) (Förstemann et al., 2005; Okamura et al., 2008a, 2008b; Saito et al., 2005). Curiously, distinct Loqs isoforms play specific roles: Loqs-PB binds Dcr-1 to promote miRNA biogenesis, while Loqs-PD binds Dcr-2 and is essential for siRNA production (Hartig et al., 2009; Zhou et al., 2009). Notably, two hpRNAs exhibit extensive complementarity to protein-coding genes (Czech et al., 2008; Okamura et al., 2008b).

In this study, we expand the annotation of Drosophila hpRNAs, validate their common biogenesis via the RNAi machinery, and demonstrate prominent accumulation of their siRNAs in testes. We describe the knockout of an hpRNA locus, hpRNA1, and show that endogenous repression of its highly complementary target ATP synthase-β in the testis is required for normal male fertility. Indeed, we find that all hpRNAs exhibit extensive complementarity to specific protein-coding genes. Importantly, although all D. melanogaster hpRNAs emerged recently during Drosophilid evolution, we identify clear selection signatures that maintain target pairing. Therefore, hpRNAs evolve adaptively to regulate testis gene expression. Finally, we elucidate first substantial phenotypic defects in RNAi mutants, which exhibit severely compromised male fertility and defective sperm development, consistent with loss of the testis-directed hpRNA pathway.

RESULTS

Drosophila hpRNA-Derived siRNAs Exhibit Testis Preference as a Class

To date, only two hpRNA loci have been experimentally validated (hp-CG4068 and hp-CG18854), but we proposed additional candidates (Okamura et al., 2008b). By assessing aggregated small RNA data, we find that hp-CG18854, hp-CG4068, the three pncr009 family loci (hp-pncr009, hp-CR32207, and hp-CR32205), and hpRNA1 are now strongly supported by thousands to millions of reads (Figure S1 available online; Table S1). Curiously, although hpRNAs were previously studied in S2 cells and ovaries (Okamura and Lai, 2008), comparison of hpRNA-siRNA tissue preferences revealed that all are dominantly expressed in testes (Figure 1A). hpRNAs also accumulated in libraries from mass-isolated imaginal discs. Such preparations contain larval gonads and broadly express testis small RNAs (Okamura et al., 2008a) and mRNAs (Brown et al., 2014). Indeed, quantitative PCR (qPCR) for primary hpRNA1 transcripts, whose siRNAs were restricted to testis and “disc” libraries, revealed background levels in embryos and hand-dissected imaginal discs, slightly higher levels in whole adult males, and vastly higher levels in testes (Figure 1B).

Figure 1. hpRNAs Are Biased to Testis and Processed by the RNAi Pathway.

(A) Small RNA levels in reads per million (RPM). The top row shows hpRNA loci. While some of these generate reads in multiple libraries, all are predominantly expressed in testes and in mass-isolated imaginal disc libraries that contain larval gonads.

(B) qPCR analysis of pri-hpRNA1 confirms much higher levels in testis than in whole adults or in dissected imaginal discs that are free of gonadal contamination.

(C) Other small RNA classes, including miRNAs and siRNAs derived from TEs and cis-NATs, do not exhibit the class-wide testis bias of hpRNAs.

(D) Analysis of small RNA libraries from loqs-PD mutant males (upper) and dcr-2 mutant males (lower) show that all hpRNA-derived siRNAs are strongly downregulated in both conditions, as are TE-siRNAs and cis-NAT-siRNAs. The “long” duplex canonical miRNAs mir-989 and mir-956, whose lengths overlap those of the shortest hpRNAs (e.g., hpRNA1) were not downregulated.

(E) Northern blotting of hpRNA expression constructs transfected into S2 cells demonstrates that hpRNAs can be functionally defined as siRNAs, since their bulk population in total RNA is resistant to β-elimination while the bantam miRNA is sensitive. Moreover, their resultant small RNAs are preferentially sorted to AGO2, while the bantam miRNA is preferentially sorted to AGO1.

(F) Luciferase sensor tests confirm regulatory capacity of transfected hp-mir-997 and hpRNA1 expression constructs. Data are the mean of quadruplicate assays ± SEM. See also Figures S1–S3 and Tables S1 and S2.

This dominant tissue preference of hpRNA-siRNAs arises from a combination of mechanisms. mRNA-seq data shows that transcription of several hpRNA loci is highest or exclusive to testes (Figure S2). However, some hpRNAs are well-expressed in other tissues, suggesting that siRNA biogenesis or stability may be more robust in gonads. This is supported by the fact that TE-siRNAs and cis-NAT-siRNAs, which are broadly detected in many tissues, accumulate to highest levels in ovaries and testes (Figure 1C). Overall, we surmise that endogenous hpRNA activity is relevant in testes, a tissue little-studied with respect to fly small RNA pathways.

Biogenesis of hpRNA Loci Is Generally Dependent on the RNAi Pathway

To generalize the biogenesis mechanism established for hp-CG4068 and hp-CG18854, we used deep sequencing to evaluate how other hpRNAs were affected in RNAi mutants. We generated 64–85 million reads from wild-type Canton S (CS), dcr-2[G31R], and loqs[PD] males (Table S2). We were interested to compare hpRNAs with “long” miRNA loci, since the longest annotated miRNAs (mir-989, mir-959, and mir-997) and the shortest hpRNAs (e.g., hpRNA1) exhibit similar hairpin duplexes of 60–80 bp.

All hpRNA-siRNAs were indeed strongly downregulated in dcr-2 and loqs[PD] mutant males, relative to CS, whereas miR-956 was unchanged and miR-989 was even slightly upregulated (Figure 1D). Curiously, miR-997 behaved like an hpRNA, since it was strongly dependent on both Dcr-2 and Loqs-PD (Figure 1D), and was restricted to testes and mass-isolated discs, at both small RNA and mRNA levels (Figures 1A and S2). The original annotation of mir-997 rested on only 13 total reads (Ruby et al., 2007), of which the dominant sequence (10 reads) maps to two genomic locations. Our consideration of ~34,000 miR-997 reads revealed two mir-997 progenitor hairpins in the Drosophila genome. Both exhibit hairpin structures to which some mature reads map uniquely, indicating their independent transcription and processing (Figure S1A). These observations support their reclassification as hpRNAs (hp-mir-997-1 and hp-mir-997-2).

Since the eight hpRNA loci behaved similarly, we compared their aggregate properties with other small RNA classes. We observed that miRNAs were not affected by mutation of dcr-2 or loqs-PD, whereas TE-siRNAs and cis-NAT-siRNAs were depleted similarly to hpRNA-siRNAs (Figure 1D). Although the expression of hpRNA1 and hp-mir-997 was not sufficiently high to judge their modulation in deep sequencing data sets outside of testes, extant ovary and/or head sRNA data showed that pncr009-family loci were depleted in ago2 mutants and enriched upon oxidization and in AGO2-IPs, similar to well-established siRNAs derived from hp-CG4068, hp-CG18854, TEs, and cis-NATs (Figure S3).

We validated the genesis of endo-siRNAs from newly characterized hpRNAs using northern blotting. Endo-siRNAs resist β-elimination, owing to methylation of their 3′ termini in AGO2 complexes, whereas miRNAs loaded in AGO1 are sensitive and exhibit increased mobility (Horwich et al., 2007). As expected, bantam miRNA was highly sensitive to β-elimination, whereas hp-CG4068B was resistant (Figure 1E) (Czech et al., 2008; Okamura et al., 2008b). We further observed that the dominant cloned species derived from transfected hpRNA1, hp-CR32207, hp-pncr009, hp-mir-997-1, and hp-mir-997-2 constructs were ~21 nt in length and predominantly resistant to β -elimination (Figure 1E). Moreover, IP assays showed that small RNAs from all hpRNA constructs preferentially accumulated in AGO2, whereas bantam associated with AGO1 (Figure 1E). Finally, assays of regulatory capacity showed that a UAS-DsRed-hpRNA1 construct repressed luciferase sensors bearing target sites for either hpRNA1–5pA or hpRNA1–3pA (Figure 1F). Similar tests of hp-mir-997-1 and hp-mir-997-2 showed they could also repress a synthetic sensor (Figure 1F).

In summary, we substantially broaden previous studies by demonstrating nine hpRNA loci are processed by the Drosophila endo-siRNA pathway into Dcr-2/Loqs-PD-depen-dent endo-siRNAs that program functional AGO2 complexes.

hpRNA1 Is Derived from Its Target Gene ATP Synthase-β

We and others reported that hp-CG4068 and hp-CG18854 have endogenous highly complementary targets, namely mus308 and CG8289 (Czech et al., 2008; Okamura et al., 2008b). Interestingly, the dominant siRNA from hpRNA1, encoded on chromosome 3, exhibits 20 continuous base pairs with the coding region of ATP synthase-β, located on chromosome 4. In fact, the left arm of hpRNA1 forms >70 bp of duplex with ATP synthase-β (Figures 2A and S1B).

Figure 2. Drosophila hpRNA1 Is an Endogenous Repressor of ATP-Synthase-β.

(A) Drosophila hpRNA1 exhibits extensive complementarity to ATP-synthase-β. The most highly expressed siRNA from the 5′ arm of hpRNA1 (5pA) guides the endogenous cleavage of ATP-synthase-β, as evidenced by the dominant isolation of 5′ RACE clones that terminate between nucleotides 10 and 11 of hpRNA1–5pA. The secondary RACE product was not related to any major hpRNA1 product.

(B) RNA-seq data shows that ATP-synthase-β is expressed at lower levels in the testes relative to the rest of the animal.

(C) qPCR validation of lower ATP-synthase-β in testes.

(D) ATP-synthase-β is upregulated in dcr-2 mutant testis. Data are the mean of triplicate assays ± SEM.

As befits a metabolic gene, FlyAtlas (Chintapalli et al., 2007) and modENCODE (Brown et al., 2014) data indicated ubiquitous expression of ATP synthase-β however, its levels were substantially lower in testes and male accessory glands (Figure 2B). We used qPCR to confirm lower levels of ATP synthase-β in testes relative to whole adults (Figure 2C) and to show ATP synthase-β was elevated >2 fold in dicer-2 mutant testes (Figure 2D). Thus, the lower level of ATPsynthase-β in testes is partly due to an RNAi mechanism.

To obtain direct evidence for endogenous regulation of ATP synthase-β by hpRNA1, we searched for hpRNA1 -directed cleavage. We performed 5′ RACE using a gene-specific primer ~150 nt 3′ to the hpRNA1 -complementary region of ATP synthase-β and sequenced 25 clones. Of these, nine clones mapped 4 nt downstream of the 5′ end of hpRNA1–5pA, which did not seem to relate to hpRNA1 -derived small RNAs. However, the 5′ end of the most abundant product (ten clones) mapped precisely between positions 10 and 11 measured from the 5′ end of hpRNA1–5pA, which is the dominant siRNA product of hpRNA1 (Figure 2A). These data reveal endogenous cleavage of ATP synthase-β programmed by hpRNA1-derived endo-siRNAs in the animal.

All Drosophila hpRNAs Can Be Associated to Specific Target Genes

We sought to broaden the evidence that hpRNAs might regulate specific targets. Searches for potential targets of hp-mir-997 revealed that the coding region of CG15040 contains 19 consecutive Watson-Crick base pairs to its dominantly cloned siRNA (Figure S1A). FlyAtlas and modENCODE data demonstrate that CG15040 is specifically expressed in testes, and this 1,895 aa protein lacks homologs in most fly species and lacks homologies to the Conserved Domain Database (Marchler-Bauer et al., 2011), suggesting that it is recently evolved. We could also identify targets of the “pncr009” family of hpRNAs (hp-pncr009, hp-CR32207, and hp-CR32205) are clustered within a small interval on chromosome 3L (Figure 3A). Remarkably, the pncr009 loci exhibit clear ancestral relationships with ten protein-coding members of the 825-Oak family, also located in this genomic locus. Three features deserve note.

Figure 3. Three hpRNAs of the pncr009 Family Target Multiple Neighboring Genes.

(A) A 60 kb genomic region encompasses three related hpRNAs: hp-pncr009, hp-CR32207, and hp-CR32205 (red). These hpRNAs exhibit extensive complementarity to ten related protein-coding genes of the 825-Oak family (blue). Some of the hpRNAs are transcribed antisense to 825-Oak genes, whereas other 825-Oak genes form divergent pairs (gray highlighted). Small RNA sequencing evidence shows that the abundant siRNAs (dominant 21 to 22 nt reads) map to the hpRNA loci.

(B) Zoomed view of hp-pncr009 and its antisense target CG32212.

(C) Example of an siRNA that is commonly produced from two hpRNAs, which exhibits full complementarity to six different 825-Oak genes.

(D) Phylogenetic tree shows that the left arms of all three hpRNAs cluster together with the mRNA targets, while their right arms form a separate clade.

(E) Luciferase sensor assays demonstrate that multiple pncr009 hairpins can repress multiple genes of the 825-Oak family. Data are the mean of quadruplicate assays ± SEM. See also Figures S1 and S4.

First, all three hpRNAs are transcribed antisense to individual 825-Oak genes. Thus, hp-pncr009 is antisense to CG32212, hp-CR32205 is antisense to CG33255, and hp-CR32207 is anti-sense of CR42842 (note that while the latter is designated as a ncRNA, it carries an open reading frame that is clearly homologous to other 825-Oak members) (Figure S4). In all cases, the left hairpin arms overlap the coding portions on the other strand (e.g., Figure 3B) with flanking pseudogenized fragments that form their hairpin stems. However, there is also extensive homology of pncr009 family-derived siRNAs to other 825-Oak family genes, with some siRNAs bearing full complementarity to targets in trans (Figure 3C). Second, three pairs of 825-Oak protein-coding genes (825-Oak:CG12519, CG14096:CG32214, and CG18294:CG32213) are arranged as closely apposed and divergently transcribed gene pairs (Figure 3A). We infer that transcription across such inverted tandem duplicates could have engendered pncr009-family hpRNAs. Third, we infer that only a single arm of each pncr009-family hpRNA is selected for regulation of 825-Oak family transcripts, since all of the hpRNA left arms (and none of their right arms) cluster with, and are complementary to, these target genes (Figure 3D). We verified the regulatory capacity of multiple pncr009 family members on several members of the 825-Oak family using luciferase sensor assays (Figure 3E).

In summary, Drosophila hpRNAs comprise an unprecedented class of metazoan endo-siRNAs linked to specific, highly complementary, trans-encoded target transcripts.

hpRNA:Target Pairings Are Subject to Adaptive Evolutionary Selection

The existence of highly conserved miRNA seed matches provides evidence for purifying selection on regulatory interactions (Bartel, 2009). However, as such information is lacking with non-conserved miRNA:target pairings, it is unclear whether newly emerged miRNA sites or miRNA loci are functionally beneficial. Of note, all identified hpRNAs emerged recently, within the melanogaster subgroup of Drosophilids. Thus, the oldest hpRNAs arose 6–8 million years ago (MYA) and the youngest 2–3 MYA, timescales for which it is impossible to evaluate evolutionary selection on putative miRNA pairings. However, the extensive target pairing of hpRNAs provided a platform to judge potential coevolution of hpRNA-derived endo-siRNAs with their targets.

We first validated that the hpRNAs under consideration are functionally processed in the species range inferred from genome alignments. Analysis of male body small RNA data from multiple close relatives of D. melanogaster (D. simulans, D. sechellia, D. yakuba, and D. erecta) recovered small RNA reads indicativeofsiRNAs from all hpRNAs (Table S3). Except for D.yakuba male body data, which did not contain hpRNA1, all loci generated 21 to 22 nt dominant small RNAs from their orthologous hpRNAs (Figure S5). Therefore, we observe functionally conserved processing of evolutionarily nascent hpRNA loci.

We proceeded with detailed selection analyses. In particular, we sought evidence of compensatory evolution, which would indicate adaptive function of species-restricted siRNA-target relationships. The properties of hpRNA1 and ATP synthase-β are illustrative. Although ATP synthase-β is broadly conserved, hpRNA1 is present in a single Drosophild subtree (Figure S1B). Both siRNA and target sequences have diverged somewhat among these species, with more mutations sustained in the hpRNAs. The target nucleotide changes were mostly silent changes at wobble positions, whereas both hairpin arms of hpRNA1 incurred changes throughout their length. We performed statistical tests to ask whether the mutations incurred on the guide and passenger arms of hpRNAs exhibited preference for coevolution for hairpin pairing and target pairing (see Experimental Procedures).

Figure 4A highlights compensatory substitution evidence for the hpRNA1 structure, p = 0.02 (Figure S1B; Table S4). To control for hairpin pairing coevolution, we randomly reordered the guide and passenger arms of hpRNAs within the multiple species alignments. Analysis of these shuffles showed no statistically significant compensatory substitution evidence, demonstrating specificity of our analysis (Table S5). Therefore, unlike bulk canonical pre-miRNA hairpins (Okamura et al., 2008d), hpRNAs evolve via compensatory hairpin arm divergence.

Figure 4. Abundant Evidence for Compensatory Substitutions that Maintain Pairing of hpRNA Arms and of hpRNAs to their Targets.

(A–C) Multiple alignments with compensatory substitution changes between the siRNA:passenger and siRNA:target are shown for (A) hpRNA1, (B) hp-CR32207, hp-CR32205, hp-pncr009, and (C) hp-mir-997. The highlighted compensatory substitution changes are relative to the reference species (dm). Three-way compensatory substitution is defined as (1) both siRNA:target and siRNA:passenger undergo double substitutions on the same base (dark green) or (2) siRNA:target undergoes a double substitution (dark green) and siRNA:passenger undergoes a single substitution on the same base (dark blue). Color legend for substitutions in the multiple alignments is shown on the right. Yellow bar highlights the columns where siRNA compensatory changes show adaptation to target change. Aside from the three-way compensatory changes highlighted by yellow bars, there are also cases where siRNA:target undergoes a double substitution and siRNA:passenger undergoes a double or single substitution on the same base, but it involves an unpaired base in the reference species. We do not consider such cases as strong three-way compensatory changes, but they do provide supporting evidence for maintenance of the structure and siRNA adaptation to target changes (highlighted by yellow bar in siRNA:target alignment only). See also Figure S1 and Tables S5.

We next tested for coevolution of siRNAs with target. Interestingly, the dominant hpRNA1-derived siRNA exhibited strong evidence for compensatory pairing with ATP-synthase β (Figure 4A), while the less-expressed siRNA from the hpRNA1 guide arm did not (Figure S1B; Table S4). This highlights that extensive pairing alone does not guarantee evolutionary selection for utility. We performed additional controls by analyzing noncognate hpRNA1/ATP-synthase β species pairs, as well as by comparing hpRNA1 sequences to ATP-synthase β orthologs from genomes that lack endogenous hpRNA1. All of these tests also failed to exhibit compensatory pairing (p>>0.05; Table S5). Overall, these tests provide compelling evidence that, in spite of its recent evolutionary emergence, adaptive selection acts upon hpRNA1:ATP synthase-β pairing to constrain this regulatory interaction.

We performed similar analyses of other hpRNAs and proposed targets. All of these hpRNAs emerged much more recently than hpRNA1, with most of them conserved only in D. simulans and/or D. sechellia. This limited our statistical power to evaluate compensatory changes. Nevertheless, most hpRNAs (except for pncr009 family loci) showed an excess of compensatory mutations (Figure 4; Table S4), revealing this as a general property of hpRNA duplex evolution. More importantly, these tests also yielded statistically significant evidence for evolutionary pressure of specific siRNAs to coevolve with their target genes (Figure 4). For these analyses, we excluded direct sense/antisense-transcribed pncr009-family siRNA/target pairs, for which all nucleotide changes were by definition “compensatory” (we thus excluded trans-targeted pairs for which an identical siRNA could be mapped to a sense/antisense pair). Examples of clearly covarying endo-siRNA and target sequences, by virtue of double compensatory changes, are highlighted yellow in Figure 4, and additional documentation and statistics are in Table S4.

Remarkably, we even observed evidence for coordinated, compensatory divergences involving target, siRNA, and passenger nucleotides (Figure 4, bolded+yellow highlighted positions). These three-way double compensatory patterns were most striking with hpRNA1/ATP synthase β and hp-mir-997/ CG15040 (Figures 4A and 4B). We infer that such cases reflect a chain of events from target divergence, to siRNA adaptation, and then to passenger strand compensation, a series of linked changes that could only be driven by selection for biological utility. Altogether, these data are notable as they represent not only first evidence for evolutionary selection for endo-siRNA:target regulation, but also because they were analyzed within a phylogenomic spread for which it is not possible to infer selection upon miRNA:target regulation.

An hpRNA1 Deletion Mutant Derepresses ATP Synthase-β and Impairs Male Fertility

Although adaptive evolution of hpRNAs with their targets is a strong reflection of beneficial regulatory interactions, dcr-2 and ago2 mutants were reported as fairly normal (Lee et al., 2004; Okamura et al., 2004). Since these mutants compromise thousands of siRNA loci, loss of individual siRNA loci might not be expected to be noticeable. Still, we considered hpRNA1 a compelling candidate for mutant analysis, given its highly specific expression and target.

We generated an hpRNA1 mutant using a Minos insertion in CG4770. Screening of ~2,000 candidate excisions yielded a lethal 4,064 nt deletion removing portions of CG4770, CG4462, and all of hpRNA1 (Figure 5A). We generated two transgenes that restore the neighboring protein-coding genes, one bearing an 8.6 kb genomic fragment and another lacking the 150 nt hpRNA1 hairpin. Both transgenes fully rescued the lethality of the deletion, indicating that hpRNA1 is not essential for viability. We refer to flies bearing the deletion covered by the genomic rescue lacking hpRNA1 as “hpRNA1 mutants” and used the deletion complemented by the complete genomic rescue as reference “hpRNA1 rescue” flies.

Figure 5. An hpRNA1 Mutant Deregulates ATP synthase-β and Impairs Male Fertility, as Do RNAi Pathway Mutants.

(A) A Minos element in CG4770 was mobilized to yield a deletion of two protein-coding genes and hpRNA1. We introduced an 8.6 kb genomic rescue fragment into this background, which covers the deleted loci and extends into both flanking genes. Alternatively, we introduced a similar genomic fragment in which hpRNA1 was deleted. We used the latter as a hpRNA1 mutant combination and the former as a wild-type comparison.

(B) Loss of hpRNA1 causes upregulation of ATP synthase-β.

(C) Male fertility assays. To exhaust their sperm capacity, we mated individual males to three females, serially transferred these every other day for 2 weeks, and counted all progeny (n > 50 males all genotypes). Hemizygous loss of hpRNA1 reduced male fertility, which was restored by the hpRNA1 genomic locus. Hemizygous null mutants of dcr-2 or ago2 also caused strong fertility defects, which were also rescued by cognate genomic trans-genes. Data are mean ± SEM. For each mutant condition, Wilcoxon rank-sum tests were done to compare them to [w1118] (gray) or to their cognate genomic rescue (blue); ***p < 0.0005.

(D–G) Adult eyes that carry ey-Gal4 GMR-Gal4, and UAS-dcr-2 transgenes, with or without other responder transgenes.

(D) The tester stock alone exhibits a normal eye.

(E) Inclusion of UAS-hpRNA1 nearly obliterates the eye.

(F) Coexpression of UAS-hpRNA1 and UAS-ATP-synthase β strongly rescues the eye.

(G) Coexpression of UAS-hpRNA1 with a UAS-ATP-synthase β construct mutated for the hpRNA1-complementary region (see below) fully rescues the eye. See also Figure S6.

We compared the expression of ATP synthase-β between these genetic backgrounds. Analysis of independent RNA samples from manually dissected tests preps showed ~2.5-fold up-regulation of ATP synthase-β in hpRNA1 mutants (Figure 5B). This serves as direct evidence for endogenous regulation of a target gene by a trans-encoded siRNA locus and shows that this locus accounts for ATP synthase-β upregulation in dcr-2 mutant testes (Figure 2D). Given the testis-specific deployment of hpRNA1, we assayed fertility of mutant and rescued males. To exclude possible second-site effects, we tested these in trans to an independent chromosomal deficiency. We observed that hpRNA1 mutant/Df males sired significantly fewer progeny compared to wild-type males and, importantly, that reintroduction of a genomic copy of hpRNA1 substantially rescued their fertility (Figure 5C).

We used gain-of-function experiments to assign the phenotypic effects of hpRNA1 to misregulation of ATP synthase-β. We did not observe substantial effects of misexpressing hpRNA1 using the well-characterized germ cell driver nanos-Gal4 (data not shown), likely due to restricted activity of the Gal4-UAS system in testes (White-Cooper, 2012). Misexpression of hpRNA1 in discs also did not induce substantial patterning defects (data not shown). However, inspired by the fact that synthetic hairpin siRNA constructs are potentiated by Dcr-2 (Dietzl et al., 2007), we coexpressed hpRNA1 with Dcr-2 in the developing eye and observed that this obliterated the adult eye (Figures 5D and 5E). This was attributable to repression of ATP-synthase-β, since coexpression of ATP-syn-thase-β almost completely ameliorated the effects of ectopic hpRNA1, leaving only mild ommatidial irregularity (Figure 5F). As the wild-type ATP-synthase-β construct is still subject to repression by hpRNA1, we generated a companion transgene bearing silent compensatory mutations in the hpRNA1-comple-mentary region that maintained its coding sequence. We found that coexpression of the ATP-synthase-β-mut construct fully suppressed the destructive capacity of ectopic hpRNA1 on eye development (Figure 5G). These in vivo genetic assays support ATP-synthase-β as the major, if not sole, phenotypic target of hpRNA1.

Male-Specific Impact of RNAi Mutants on Fertility

Our data on adaptive regulation of testis gene expression by the hpRNA pathway led us to re-evaluate the phenotypic impact of Drosophila RNAi mutants. Endo-siRNAs are elevated in gonads relative to somatic tissues (Brown et al., 2014; Wen et al., 2014). However, while both female and male gonads express TE-siRNAs and cis-NAT-siRNAs, the testis is distinguished by its accumulation of hpRNA-siRNAs (Figure 1A). We therefore asked if there were any sex-biased phenotypes of RNAi mutants.

We did not observe any defects in female fertility of RNAi mutants relative to their rescued counterparts (Figure S6). In contrast, assays of the null alleles dcr-2[L811fsx] and ago2 [454] in trans to deficiencies showed they were severely compromised for individual male fertility (Figure 5C), much more so than hpRNA1. The defects of both RNAi mutants were rescued by introduction of single copies of the cognate genomic transgenes. Thus, more than a decade following the initial description of these mutants, we reveal an unexpected male-specific defect in RNAi mutants that correlates with the testis-biased hpRNA pathway.

hpRNA1 and RNAi Mutants Affect Spermatogenesis

With knowledge of a testis-specific RNAi pathway that is required for robust male fertility, we sought potential defects in spermatogenesis. During sperm development, mitotic and meiotic divisions of male germline cells, each with incomplete cytokinesis, generate groups of 64 haploid round spermatids. These are initially interconnected but separate at the end of elongation via the coordinated movements of actin-based investment cones (ICs) that form over the mature, needle-shaped nuclei. The ICs are well-aligned within the cystic bulge of an individualization complex, which extrudes the cytoplasmic contents of the spermatids to form 64 individual sperm that are then stored in the seminal vesicle (Fabian and Brill, 2012).

Consistent with a defect in sperm development, seminal vesicles from hpRNA1 mutant/Df males contained reduced numbers of mature sperm relative to wild-type, a phenotype rescued by presence of the hpRNA1 rescue transgene (Figures 6A–6F). To gain insight into the nature of the sper-matogenesis defect, we analyzed ICs and nuclei by labeling testes with rhodamine phalloidin (for ICs), antibodies against myosin VI, and DAPI (for DNA). Overall nuclear organization in the testis was mildly defective (Figures 6G–6I), including poorly aligned nuclei revealed at higher magnification (Figures 7A–S7C). More significantly, we observed a penetrant defect in IC alignment within the cystic bulges of hpRNA1/Df mutants, and this was rescued by the hpRNA1 rescue transgene (Figures 6J–6L; see also Figure S7D and S7E). Therefore, deletion of a single hpRNA locus, despite its seemingly modest level of expression (Figure 1), has detectable impacts not only on testis gene expression and male fertility but also on sperm development.

Figure 6. The hpRNA/RNAi Pathway Is Required for Spermatogenesis.

Stainings of various regions of the male reproductive apparatus. Top two rows, seminal vesicle shown at low and high magnification; middle two rows, whole testis; bottom two rows, cystic bulges with individualization complexes. Scale bars are noted.

(A and B) Wild-type seminal vesicle filled with needle-shaped sperm nuclei, as stained with DAPI.

(C and D) hpRNA1 mutant seminal shows lower density of sperm, which is rescued by hpRNA1 locus ([E] and [F]).

(G–I) Wild-type testis exhibits groups of individualization complexes with ordered nuclei and associated actin cones (arrowhead). These are mostly normally organized in hpRNA1 mutant (H) or rescued (I) testis.

(J) High magnification of individual wild-type cystic bulge shows well-aligned ICs composed of actin (bracket).

(K and L) (K)hpRNA1 mutant cystic bulge exhibits lagging ICs (asterisks), and this is fully rescuable (L).

(M–T) Both null RNAi mutants, dcr-2 and ago2, are severely depleted for sperm in the seminal vesicle.

(U and V) dcr-2 mutants exhibit highly dispersed spermatid nuclei.

(W and X) ago2 mutants exhibit severely disorganized spermatid nuclei, such that it is difficult to identify individualization complexes.

(Y and Z) dcr-2 mutants exhibit individualization complexes with reduced numbers and poorly aligned ICs ([Y’], bracket). About a third of cases exhibit even more dispersed ICs that are difficult to assign to cystic bulges.

(AA) ago2 mutants exhibit highly dispersed ICs that do not organize into identifiable cystic bulges; this is rescued by ago2 transgene (BB). See also Figure S7.

Figure 7. Model for the Emergence and Evolutionary Maturation of Drosophila hpRNAs.

(A) In an early stage of this process, two protein-coding genes in inverted orientation are subject to read-through transcription yielding a long inverted repeat. The resulting hairpin generates siRNAs that are self-complementary to the mRNAs.

(A’) An alternate birth stage could be a protein-coding gene, which has an antisense hairpin-forming transcript, whose siRNAs are necessarily complementary. This arrangement is observed for each of the three pncr009 family hpRNAs, which are transcribed antisense to individual 825-Oak family members.

(B) Subsequently, a distinct hairpin transcript may be retained in proximity to, and have extended complementarity to, the progenitor mRNAs. This state is reflected in the genomic clustering of pncr009 family hpRNAs and other targets of the 825-Oak family, many of which are still arranged in divergent orientations.

(C) In the next step, the hpRNA and target become genomically separated yet retain extensive complementarity to each other. The hp-CG18854/CG8289 and hpRNA1/ATP synthase-β pairs are representative of this state.

(D) Finally, in the mature state of this process, the hpRNA/target interaction has been whittled to a specific siRNA that is highly complementary to its target. This situation is exemplified by the hp-CG4068/mus308 interaction.

(E) At various stages in the lifecycle, hpRNA amplifications are observed, either in cis or in distinct genomic copies. These additional copies may permit the aggregate loci to sample greater mutational space that facilitates target adaptation.

Encouraged by these results, we analyzed RNAi mutants. Both hemizygous dcr-2[L811fsx] and ago2[454] mutants exhibited dramatically reduced numbers of sperm in their seminal vesicles (Figures 6M–6T). The depletion of sperm in ago2 mutants was stronger than in dcr-2, and this correlated with the lower fertility of ago2 mutant males (Figure 5C). We confirmed the stronger sperm loss of ago2 in another trans-heterozygous null allelic combination ago2[321]/Df (Figures S7H–S7K). The reason for the phenotypic discrepancy of dcr-2 and ago2 is not clear, but there are precedents where Ago effector mutants exhibit stronger effects compared to upstream pathway components (Halic and Moazed, 2010; Smibert et al., 2013). In any case, all dcr-2 and ago2 sperm deficiencies were fully rescued by the cognate genomic transgenes, indicating these are genuine phenotypes.

The RNAi mutants showed IC and nuclear alignment defects akin to those in hpRNA1 mutants. However, consistent with the notion that RNAi mutants remove bulk siRNA function, including of hpRNAs in general, these phenotypes of dcr-2 and ago2 mutants were substantially more pronounced. We observed scattered nuclei in dcr-2 (Figure 6U, compare to rescue in Figure 6V). These defects were even stronger in both ago2 mutants analyzed, such that the normal clusters of spermatid nuclei could rarely be seen (Figure 6W, compare to rescue in Figure 6X; see also Figures S7L and S7M). This effect carried over to the cystic bulge, such that dcr-2 exhibited scattered and reduced numbers of ICs (Figures 6Y and 6Z; see also Figures S7F and S7G), whereas individualization complexes were so disorganized in ago2 that it was difficult to identify overt cystic bulges (Figures 6AA, 6BB, S7N, and S7O). Overall, such defects in IC and nuclear organization are observed in other subfertile Drosophila mutants (Fabrizio et al., 1998) and can explain the observed defect in sperm production.

We conclude that Drosophila RNAi mutants, although previously suggested to exhibit only mild and general defects in gene expression, actually reveal substantial male-specific defects that relate to sperm differentiation and negatively impact male fertility. These findings are consistent with a general loss of the testis-restricted hpRNA pathway (Figure 1) in RNAi pathway mutants.

DISCUSSION

Distinct Evolution and Function of Hairpin RNAs and miRNAs

Drosophila hpRNAs and miRNAs both use inverted repeat precursor transcripts to generate short regulatory RNAs (Okamura et al., 2008c). However, they have distinct biochemical and functional properties, due to association of miRNAs with AGO1 (permitting repression of modestly paired targets) and loading of siRNAs with AGO2 (ensuring on-target slicing of highly complementary targets). These characteristics are married to distinct evolutionary properties. Recently emerged miRNAs are expressed at low levels and believed to have few, if any, beneficial targets (Bartel and Chen, 2004; Chen and Rajewsky, 2007). In contrast, all hpRNAs evolved recently, yet they achieve substantial levels of expression. In the case of hpRNA1, we explicitly show its deletion impacts endogenous target gene expression and male fertility, and we further show the RNAi pathway has profound male-specific requirements for gametogenesis and fertility.

Importantly, we show that target pairing of hpRNA-siRNAs is broadly subject to adaptive evolutionary selection. We further note the phylogenetic range within which we detect statistical evidence of compensatory substitutions between hpRNA-derived siRNAs and their targets is insufficient to obtain evidence for target selection of recently evolved miRNAs. Therefore, while only miRNAs that are well-conserved seem to have substantial impact on gene expression, the raison d’être of hpRNAs seems precisely to mediate species-specific gene expression.

Control of Gonadal Gene Expression by the hpRNA Pathway

It is striking that transcription and/or siRNA-generating potential of all hpRNA loci is elevated in gonads and often restricted to testes. Among settings and conditions conducive to gene emergence and rapid gene evolution (Chen et al., 2013), the testis is a particularly facile environment (Zhao et al., 2014). Here, diverse genetic conflicts play out, including selfish genes, speciation, maternal/paternal conflicts, and genomic conflicts. With respect to the latter, the Nmy-Dox system is relevant. The recently evolved Dox gene impairs development of Y-bearing sperm in D. simulans and has consequently been neutralized by the inverted repeat locus Nmy, which bears homology to Dox (Tao et al., 2007a; Tao et al., 2007b). Interestingly, the secondary structure of Nmy and its regulation of Dox resemble features that we characterized for hpRNAs and their targets in D. melanogaster, even though this species lacks Dox or Nmy genes.

As we did not observe sex-ratio defects in hpRNA1, dcr-2, or ago2 mutants (not shown), the hpRNA pathway may not be dedicated to genomic conflicts that threaten male progeny. On the other hand, sex-ratio systems evolve recurrently throughout the Drosophilid phylogeny and can have extraordinarily rapid dynamics, with evidence for a selective sweep of sex-ratio chromosomes followed by their suppression within 100 years (Bastide et al., 2011). It may be that D. melanogaster does not currently experience a strong need for suppression of sex-ratio distorters by hpRNAs, even though its closest relatives do. Nevertheless, some D. melanogaster hpRNA targets are recently evolved and reside on the X chromosome (e.g., the mir-997 target CG15040), and their potential involvement in genomic conflict deserves further study.

In cases such as hpRNA1:ATP synthase-β, the hpRNA target is well-conserved and not obviously a rogue protein. Still, analysis of hpRNA1 knockouts shows this regulatory interaction is required for normal spermatogenesis. We note that sphinx is a recently evolved ncRNA derived from ATP synthase chain F and has a male-specific transcript involved in courtship behavior (Dai et al., 2008; Wang et al., 2002). Curiously, sphinx is located only 50 kb from ATP synthase-β, which we showed is the target of hpRNA1. Therefore, structural genes for ATP synthase have engendered de novo protein-coding and non-coding RNA genes. This hints at an evolutionary underpinning to gene flux involving the core ATP-generating machinery in sperm.

Drosophila hpRNAs and Plant miRNAs Emerge via Similar Mechanisms

Our expanded analysis of Drosophila hpRNA genes shows that hpRNAs generally derive from their target genes. This parallels the notion that many plant miRNAs similarly evolved from their targets (Allen et al., 2004), a phenomenon recently observed in Cnidaria (Moran et al., 2014). Notably, the collection of identified Drosophila hpRNAs comprise likely representatives along a timeline of hpRNA gene evolution (Figure 7).

We propose the early state involves local genomic duplications. In the case of the pncr009 supercluster, a history of duplications of 825-Oak family genes resulted in multiple members of both protein-coding and hpRNA genes. The pncr009 hpRNAs may have been born through antisense transcripts that picked up neofunctionalized duplicate fragments or through read-through transcripts across divergent gene pairs. Following local duplications, hpRNAs may become genomically unlinked from their targets but still retain signatures of extended homology. This situation applies to hpRNA1:ATP synthase-β and hp-CG18854:CG8289. Finally, we hypothesize that particular siRNAs from hpRNAs may become preferentially selected for repression, resulting in loss of extended homology between hpRNA and target. This is exemplified by hp-CG4068:mus308, where only one of many siRNAs is complementary to the target.

Despite remarkable evolutionary convergence of plant miRNA and Drosophila hpRNA pathways, there is a core distinction in their behavior that is linked to their function. Plant miRNAs and targets evolve mostly by purifying selection, leading to a substantial class of deeply conserved relationships (Floyd and Bowman, 2004), whereas Drosophila hpRNAs evolve rapidly and adaptively with their targets. Notably, intertwined with the evolutionary series we propose is the capacity for hpRNA duplications, which are observed frequently (Figure 7E). The existence of multiple homologous hpRNAs that target similar transcripts may favor the ability of each family to acquire a member that has mutated to permit covariation with their targets. An analogous mechanism allows the poxvirus antihost factor K3L to avoid the antiviral protein kinase R (PKR), whereby transient amplifications of K3L genes permit the emergence of mutants that evade PKR (Elde et al., 2012). Subsequently, these “molecular accordions” contract once an advantageous adaptation has been selected. Indeed, the 20 tandem copies of the hp-CG4068 hairpin may represent such a molecular accordion, but one that exists in the noncoding RNA world to facilitate adaptations to its target transcript.

EXPERIMENTAL PROCEDURES

Small RNA Analyses

We cloned small RNA libraries as described in Berezikov et al. (2011) from Canton S males, dcr-2[G31R] homozygous males (Lee et al., 2004), and loqs [KO] homozygous males bearing a loqs[PB] transgene (i.e., “loqs[PD]” mutants) (Hartig et al., 2009). The raw data were deposited in the NCBI Short Read Archive (SRP043164). We recently reported small RNA male body libraries for D. sechellia, D simulans, and D. erecta, D.yakuba (Mohammed et al., 2014) and used data deposited at the NCBI Gene Expression Omnibus (GSE56244). We processed these data sets and other GEO/SRA small RNA data and mapped them to the dm3 genome as described in Berezikov et al. (2011). Statistics of library sequencing and mapping are provided in Tables S1 and S2.

We mapped reads to the stems of hpRNAs (excluding loops) and 21/22 nt reads were were used in expression analysis. For TE-siRNA and cis-NAT-siRNA loci, we used our recently reported annotations (Wen et al., 2014). Only 21 nt dominant loci were considered for each library, and 21 nt reads were used in expression analysis, while all read sizes were considered for miRNAs. The tissue libraries selected for aggregated analysis for siRNAs in Figure 1 were all from wild-type genotypes.

To calculate fold changes between mutant and wild-type, we applied a filter requiring ≥20 reads and ≥10 reads to exclude lowly expressed loci for each wild-typelibraryforall mappedreadsand uniquelymapped reads,respectively.

Drosophila Mutants and Transgenes

For phenotypic analysis we used (1) the null allele dcr-2[L811fsX] (Lee et al., 2004) in trans to Df[BSC45] (BL#130353), compared to rescue with a dcr-2 fos-mid (Kemp et al., 2013), and (2) the null alleles ago2[321] and ago2[454] (Hain et al., 2010) in trans to Df[BSC558] (BL#25120), compared to rescues with a Flag-HA-AGO2 genomic transgene (Czech et al., 2008). To analyze hpRNA1, we generated a deletion of CG4770/hpRNA1/CG4462 by mobilizing Minos [MB01974]. We rescued this using transgenes that fully covered the deleted genes or that contained a specific deletion of hpRNA1 and analyzed both in trans to Df[ED6025] (BL#8964). We also made inducible transgenes for UAS-DsRed-hpRNA1 and wild-type and mutant UAS-ATP synthase β. Primers for screening imprecise excisions and cloning transgenes are provided in the Supplemental Information, as are detailed information on immunostaining procedures and fertility assays.

Compensatory Single/Double Substitution p Value Estimation Using Phylogenetic Tree

The Evofam framework estimates the statistical significance of double substitutions (Parker et al., 2011), and we extended this to include single substitution evidence. We used a phylogenetic tree to reconstruct ancestral sequences (internal nodes) using PAML (Yang, 2007). Counts of compensatory double substitutions and single compatible mutations in the tree (compensatory double substitution is defined as mutations occurring inmatching stem base positions and that occur at the same time – i.e., on the same branch of the tree, and that maintain Watson-Crick pairing) were computed. Monte-Carlo simulation is used to estimate a p value for how unlikely it is to see the measured number of compensatory single and double substitutions by comparing with those in randomly generated sets of mutations of the same size. p values were computed for compensatory evidence between hpRNA passenger and guide arms and a control with shuffled passenger:guide pairings.

To predict targets, we BLAST ed all siRNA registers to the genome and realigned all siRNA registers to hit regions (adding flanks) by Smith-Waterman local alignment, with a modified matching matrix to include G-U pairings and retrieve siRNA:target alignment scores ≥70. We computed the p value as described above between the siRNA and target, and also over an alignment extended on both sides by Smith-Waterman local alignment (including G-U pairings). Controls used were as follows: (i) permute target species paired to siRNA and (ii) target species randomly assigned from orthologs not having hpRNA. For each siRNA:target control pair we computed the p value as described above, and the shuffling was repeated ten times and the p values averaged.

For hpRNAs exhibiting three-way compensatory changes involving both hairpin arms and target divergences (hpRNA1 and hp-mir-997-1/2), we performed statistical tests to estimate likelihoods of compensatory mutations in both guide:passenger and guide:target co-occurring by chance. Due to complications in modeling potential three-way evolutionary processes, we used a simpler permutation test in which columns of siRNA:passenger and siRNA:target alignments (in Figure 4) were independently randomly shuffled 1,000 times and the probability of overlap relative to observed data was calculated. The p values obtained for hpRNA1 and hp-mir-99/27-½ with their targets were 0.0374 and 0.0486, respectively. Note that this statistical framework has less power than the Monte Carlo p value estimation using the phylogenetic tree used in two-way comparisons, so these three-way p values are clear minimum estimates.