Germline gene de-silencing by a transposon insertion is triggered by an altered landscape of local piRNA biogenesis

Transposable elements (TE) are selfish genetic elements that can cause harmful mutations. In Drosophila, it has been estimated that half of all spontaneous visible marker phenotypes are mutations caused by TE insertions. Because of the harm posed by TEs, eukaryotes have evolved systems of small RNA-based genome defense to limit transposition. However, as in all immune systems, there is a cost of autoimmunity and small RNA-based systems that silence TEs can inadvertently silence genes flanking TE insertions. In a screen for essential meiotic genes in Drosophila melanogaster, a truncated Doc retrotransposon within a neighboring gene was found to trigger the germline silencing of ald, the Drosophila Mps1 homolog, a gene essential for meiosis. A subsequent screen for modifiers of this silencing identified a new insertion of a Hobo DNA transposon in the same neighboring gene. Here we describe how the original Doc insertion triggers flanking piRNA biogenesis and local gene silencing and how the additional Hobo insertion leads to de-silencing by reducing flanking piRNA biogenesis triggered by the original Doc insertion. These results support a model of TE-mediated silencing by piRNA biogenesis in cis that depends on local determinants of transcription. This may explain complex patterns of off-target gene silencing triggered by TEs within populations and in the laboratory. It also provides a mechanism of sign epistasis among TE insertions. Author Summary Transposable elements (TEs) are selfish DNA elements that can move through genomes and cause mutation. In some species, the vast majority of DNA is composed of this form of selfish DNA. Because TEs can be harmful, systems of genome immunity based on small RNA have evolved to limit the movement of TEs. However, like all systems of immunity, it can be challenging for the host to distinguish self from non-self. Thus, TE insertions occasionally cause the small RNA silencing machinery to turn off the expression of critical genes. The rules by which this inadvertent form of autoimmunity causes gene silencing are not well understood. In this article, we describe a phenomenon whereby a TE insertion, rather than silencing a nearby gene, rescues the silencing of a gene caused by another TE insertion. This reveals a mode of TE interaction via small RNA silencing that may be important for understanding how TEs exert their effects on gene expression in populations and across species.


Introduction
Transposable elements (TE) are harmful selfish elements that can cause DNA damage, mutation, chromosome rearrangements, and sterility. Due to their capacity to cause mutation, it has been estimated that about half of the spontaneous mutations that cause visible phenotypes in Drosophila are caused by TE insertions [1] . Nonetheless, despite their harm, TEs can greatly proliferate in the genomes of sexually reproducing species [2] . A consequence of TE proliferation is that diverse systems of genome defense have evolved that limit transposition through DNA methylation, repressive chromatin, direct transcriptional repression, and small-RNA silencing. There is substantial cross-talk between these modes of genome defense.
For example, small RNAs generated from harmful TE transcripts can silence TEs through cytoplasmic post-transcriptional silencing but also enter the nucleus to trigger DNA methylation and transcriptional repression [3,4] .
In animals, small RNAs designated piwi-interacting RNAs (piRNAs) play a critical role in genome defense within reproductive tissues. piRNAs are derived from TE sequences recognized by the piRNA machinery and diverted from canonical mRNA processing into a piRNA generating pathway. By shunting TE transcripts toward piRNA biogenesis, the host is able to generate a pool of antisense piRNAs that repress TEs throughout the genome.
Interestingly, like other systems of immunity, genomic immunity can be costly when the distinction between self and non-self is disrupted. For example, in Arabidopsis thaliana , selection can act against DNA-methylated TE insertions that reduce the expression of flanking genes [5] . Off-target gene silencing by systems of genome defense, and subsequent selective effects, has been observed in a variety of organisms [6][7][8][9][10][11][12][13][14][15][16][17] . However, genic silencing by flanking TEs is hardly universal within a genome. For example, in maize, the capacity to trigger the formation of flanking heterochromatin can vary significantly among TE families [13] . The cause of this variation is poorly understood.
Studies in Drosophila , where DNA methylation is absent and piRNAs are the primary line of defense against TEs, show that TE insertions can trigger the spreading of heterochromatin and transcriptional silencing of genes [16][17][18][19][20][21] . In the germline, TE insertions also have the capacity to trigger the production of piRNAs from flanking sequences [22][23][24][25] . The mechanism of flanking piRNA biogenesis that spreads from a TE insertion can be explained by a general model whereby Piwi-piRNA complexes target nascent TE transcripts [26][27][28][29] followed by recruitment of the histone methyltransferase SETDB1/Egg [30][31][32] . Upon H3K9 methylation by SETDB1, germline TE insertions may be co-transcriptionally repressed and converted to piRNA generating loci by subsequent recruitment of the HP1 paralog Rhino [25,26,33] . Recruitment of Rhino coincides with non-canonical transcription within the TE insertion and transcripts are directed into a pathway of RNA processing that lacks standard capping, splicing and polyAdenylation [25,[34][35][36] . Since non-canonical transcription can ignore TE encoded termination signals [36] and extend beyond the target TE insertion, transcripts designated for piRNA processing can yield piRNAs from genomic regions outside the TE insertion. This occurs presumably through phased piRNA biogenesis [37][38][39] , since transcripts derived from unique genomic regions flanking the TE will not be the direct target of TE-derived piRNAs that trigger ping-pong biogenesis.
In Drosophila , there is evidence that TEs with the capacity to induce flanking H3K9 methylation through piRNA targeting are deleterious due to the silencing of neighboring genes [14,15] . However, there is striking variation in the capacity for TE insertions to trigger these effects.
Across two independent strains, only about half of euchromatic insertions show a signature of locally induced H3K9 methylation [15] . Why some TEs trigger local piRNA biogenesis and/or repressive chromatin and others do not is poorly understood, though a variety of factors are known to contribute. One factor is clearly the class of TE. In maize, only some TE families appear to induce local heterochromatin formation [13] and in Drosophila , the LTR class appears to exert a stronger effect on local chromatin compared to other families [15] . Such differences may be explained by regulatory sequences embedded within the particular TE family or class.
For example, elements primarily expressed in somatic cells of the ovary trigger a greater degree of flanking H3K9 methylation in cultured ovarian somatic cells [27] . Additionally, TE insertions that lack a promoter and are thus not expressed can fail to trigger flanking piRNA biogenesis in the germline [40] .
In the absence of regulatory sequences encoded within TE insertions, the capacity for a TE fragment to nucleate local repression and piRNA biogenesis depends on the interaction between the individual insertion and the transcriptional environment. A recent investigation of flanking piRNA biogenesis triggered by transgenes showed that transcription in opposing directions (convergent transcription) may enhance conversion of TEs into standalone piRNA clusters with flanking piRNA biogenesis [24] . However, there is no general model that explains why some TEs insertions have strong effects on the expression of flanking genes while others do not.
In a genetic analysis of the Drosophila Mps1 locus, we identified TE insertions that have a complex influence on gene expression whereby insertions can either trigger local gene silencing or de-silencing. Using polyA RNA-seq and small RNA sequencing, we show how the fate of transcripts from this locus shifts between canonical mRNA processing and piRNA biogenesis in the presence of different TE insertions. This complex effect of TE insertions supports a model in which the capacity for one TE to silence flanking genes depends on local patterns of transcription that can be altered by other TE insertions. This represents a case of compensatory mutation or sign epistasis between TE insertions, whereby the harm or benefit of an allele depends on genetic background [41] .

A DNA transposon insertion rescues a retrotransposon insertion allele of Mps1
The Drosophila homolog of Mps1 , ald , is a conserved protein kinase that is a key component of the meiotic and mitotic spindle assembly checkpoint present in most organisms [42][43][44][45] . While Mps1 has both mitotic and meiotic function, the Drosophila Mps1 A15 allele only affects meiosis and is caused by a Doc non-LTR retrotransposon insertion into the 3' end of the neighboring gene alt , rather than Mps1 itself [46,47] (Figure 1). alt and Mps1 are convergently transcribed with transcripts overlapping at the 3' end, a configuration that has been proposed to enable TE insertions to trigger flanking piRNA biogenesis and local gene silencing [24] . To understand why a transposon insertion in one gene could affect the function of another gene, a genetic screen was performed to identify suppressors of the Doc Mps1 A15 allele [48] . In this screen, seven stocks were isolated that suppressed nondisjunction caused by the

Local gene silencing by a Doc insertion is ameliorated with insertion of the Hobo element
Since the Mps1 A15 allele has an effect on meiosis, but not mitosis, we determined how the two   (Figure 3). Of note, these piRNAs are essentially derived from only one strand, in the sense orientation with respect to alt transcription. Across the entire region, there is no evidence that piRNAs are generated through bidirectional transcription since piRNAs derived from one strand do not have a corresponding population derived from the alternate strand. Thus, in the absence of TE insertions, piRNAs from this region appear to be generated through the pathway that generates sense 3'UTR genic piRNAs [51] .

Figure 2. A Hobo insertion triggers germline de-silencing of two genes silenced by the
The insertion of the Doc element in Mps1 A15 homozygotes changes this pattern dramatically. In all three experiments, the Doc insertion is associated with conversion to piRNA biogenesis from both strands. This is evident in alt , where the Doc insertion is located , but extends across three genes on one side. Interestingly, while piRNAs derived from both strands are identified from  Silencing and de-silencing act zygotically in cis piRNAs that repress TEs in Drosophila are transmitted maternally and maintain continuous silencing across generations [33,[52][53][54][55] . piRNAs also have the capacity to maintain off-target gene silencing through maternal transmission [56,57] . This maternal transmission also can enable paramutation [58] . Therefore, we tested whether the silencing or de-silencing of Mps1 depended on the maternal silencing state. This was achieved through quantitative RT-PCR of   [24] . Upon conversion to a standalone dual-strand cluster, silencing of functional transcripts is likely caused by transcription being directed away from standard mRNA processing into a pathway of piRNA biogenesis [25,34,36] . In this case, flanking piRNA biogenesis from both strands can be considered a readout of genic co-transcriptional repression. How does the Hobo insertion lead to de-silencing? The Hobo insertion still retains the Hobo promoter and TE insertions near gene TSS's have been shown to block PolII recruitment to genic promoters [27] . Transcriptional repression of alt by the Hobo insertion thus likely precludes the Doc fragment from being a sufficient target for Piwi-piRNAs that can trigger conversion into a piRNA producing locus. Importantly, while the Hobo insertion leads to a substantial reduction in the abundance of both sense and antisense Mps1 piRNAs, flanking piRNA biogenesis is not completely blocked. In this case, one might expect that Mps1 may retain some repression. Nonetheless, both RNA-seq and RT-qPCR analysis reveal that the Hobo insertion completely restores the expression of Mps1 . This supports the "all-or-nothing" model whereby euchromatic TEs can trigger either weak or strong, but not intermediate, silencing [22,24] .
Strikingly, we found no evidence for maternal effects on the expression of Mps1 , either for silencing alleles or de-silencing alleles. In Drosophila , maternal effects by piRNA play an important role in TE repression. This is revealed in syndromes of hybrid dysgenesis where paternal transmission of TEs causes excessive transposition if the mother lacks a corresponding pool of germline piRNAs. Maternal effects in TE regulation also reveal differences in how piRNA source loci depend on piRNAs for either their establishment or maintenance. Functional pericentric dual-strand clusters (such as 42AB) require maternal piRNAs but depletion of Piwi in adult ovaries does not lead to loss of cluster chromatin marks [33] Therefore, dual-strand cluster chromatin can be maintained in the absence of nuclear piRNAs. In contrast, standalone transgenes that trigger flanking piRNA biogenesis require piRNA production for maintenance of Rhino, HP1 and H3K9me3 chromatin [24] . Moreover, while maternal inheritance of I-element transgenes along with a substantial pool of I-element targeting piRNAs can trigger piRNA biogenesis from flanking regions in progeny, this mode of inheritance is not associated with altered chromatin signatures [24] . Overall, it is unclear why maternal effects and paramutation triggered by piRNAs can occur for some genes and not others [56][57][58] .
The costs of gene silencing triggered by TEs have been proposed to shape the dynamics of TEs in populations [5,14] . However, TE insertions do not universally trigger flanking gene repression. In some cases, the expression of neighboring genes can be enhanced [59] . For example, an Accord LTR insertion in Drosophila melanogaster can enhance the expression of the cytochrome P450 gene Cyp6g1 and provide resistance to DDT [60] . In Drosophila simulans , a Doc insertion near Cyp6g1 has a similar effect and has been the target of positive selection [61] . Enhanced expression is attributed to the regulatory sequences carried by these elements.
Here we present a case where TE insertions can alter the germline expression of a gene in opposing ways by altering the local profile of piRNA biogenesis. Formally, this represents a case of sign epistasis. Since the Hobo insertion alone is predicted to be deleterious through alt silencing, but beneficial when combined with the Doc insertion, this satisfies the condition of sign epistasis. For this reason, TE dynamics within populations may not solely be influenced by their single effects, but also their epistatic interactions. As genomic TE density increases, the likelihood for such interactions is expected to increase.

Identification of the Hobo insertion Mps1 A15 revertant allele
From an EMS screen [48] to identify suppressors of the Mps1 A15 Doc insertion allele of ald seven stocks were identified that suppressed non-disjunction. After one round of recombination FASTQ files were aligned to release 6 of the D. melanogaster reference genome using bwa version 0.7.7-r441 [62] . SNPs and insertion/deletion polymorphisms were identified using SAMtools and BCFtools (version 0.1.19-44428cd) [63] . Transposable elements were identified as previously described [64] . Briefly, split and discordant read pairs were isolated using SAMBlaster [65] and individual reads were annotated using a BLAST search of the canonical Drosophila melanogaster transposable element database [66] . A position with multiple reads from a single TE was defined as a putative TE insertion site and was then manually analyzed.
Using this approach, the Hobo insertion was identified. Further PCR and Sanger sequencing was used to confirm structure and insertion location within alt . To ensure piRNAs were being characterized, a second experiment was performed using an altered protocol on oxidized and non-oxidized RNA, in parallel. Pooled RNA samples were split and half the sample was oxidized [68] . RNA from each sample was ligated to 3' and 5' adapters using an rRNA blocking procedure [67] and subjected to direct reverse-transcription with unique barcoded RT primers. Barcoded RT products were pooled and size selected on a 10% acrylamide gel for the appropriate size of small RNA cDNAs (18-30 nt) appended to the additional sequence added by the adapters and RT primer. Size selection was facilitated by completing the same procedure in parallel on 18 and 30 nt RNA oligonucleotides. This procedure of pooled size selection allowed all cDNA samples to be extracted under identical conditions. Full protocol is provided in Supplemental. Size selected RT products were extracted from acrylamide, subjected to 15 cycles (non-oxidized) and 18 cycles (oxidized) of PCR and sequenced. Reads were bioinformatically trimmed of adapters, unique molecular identifiers (6bp), size selected between 23 and 30 nt and miRNA/tRNA/rRNA depleted using the CLC Genomic Workbench.
From both experiments, small RNA reads were mapped with bwa to release 6.33 of the Drosophila melanogaster reference, counts analyzed with BEDTools [69] and visualized with R.
Nucleotide composition and ping-pong signatures were analyzed with the unitas package [70] .
Quantitative PCR RNA was collected from whole ovaries. Sampling was performed in sets of three whereby three daughters were sampled for each of three mothers, thus providing replication across mothers of a given genotype for a total of 9 samples per genotype/mother combination.