Taming the Turmoil Within: New Insights on the Containment of Transposable Elements

Abstract

Transposable elements (TEs) are mobile genetic parasites that can exponentially increase their genomic abundance through self-propagation. Classic theoretical papers highlighted the importance of two potentially escalating forces that oppose TE spread: regulated transposition and purifying selection. Here we review new insights into mechanisms of TE regulation and purifying selection, which reveal the remarkable foresight of these theoretical models. We further highlight emergent connections between transcriptional control enacted by small RNAs and the contribution of TE insertions to structural mutation and host-gene regulation. Finally, we call for increased comparative analysis of TE dynamics and fitness effects, as well as host control mechanisms, to reveal how interconnected forces shape the differential prevalence and distribution of TEs across the tree of life.

Containing Transposable Elements

Transposable elements (TEs) are the exemplar of selfish genes. By parasitizing host cells and replication machinery, these mobile DNA fragments produce new copies of themselves, often achieving exceptional genomic abundance. For example, >85% of the 2.9 Gb maize genome is comprised of TE insertions [1], leaving one to wonder whether the host will soon be overtaken by the ever-proliferating parasites.

The spread of TEs through genomes is potentially exponential, owing to the increase in genome-wide transposition rate that occurs as copy numbers increase over successive generations. The conventional wisdom, therefore, is that constraining TEs requires an opposing force whose strength similarly increases with TE abundance. In the early 1980s two solutions were proposed for this conundrum: copy-number-dependent transposition (see glossary), and synergistic purifying selection [2,3]. In the presence of copy-number-dependent transposition, the per-element transposition rate decreases with increasing TE copy number, slowing and ultimately halting TE spread [2,3]. By contrast, under synergistic purifying selection, each additional TE copy imposes an escalating cost on host fitness, which accelerates their removal from populations [2] (Figure 1).

Figure 1. Exponential increase in TE copy number can be constrained in two ways.

A) Copy number-dependent transposition. As copy number increases, the transposition rate decreases and TE spread can be halted. B) Synergistic purifying selection. As copy number increases, the harmful effects of individual TE insertions become increasingly amplified.

Inspired by these theoretical insights, investigations of TE dynamics sought to identify which force is more important for TE containment. The field quickly arrived at selection as the most plausible predominant force. Estimated transposition rates consistently exceeded excision and deletion rates, demonstrating that natural populations are not in transposition-excision/deletion balance [4,5]. Additionally, analytical models suggested that neither host nor TE alleles establishing copy-number-dependent transposition were likely to invade, owing to their limited selective advantage, especially with moderate levels of recombination [6]. Finally, surveys of TE insertion polymorphism revealed an excess of rare insertions, suggesting strong purifying selection [Reviewed in 7]. Ectopic recombination in particular emerged as a promising source of synergistic purifying selection, as the potential for non-allelic exchange may increase with the square of TE copy number [8–10]. Resulting theoretical and empirical studies suggest that ectopic recombination is indeed a major target of selection against TE insertions [11–15]. However, clear evidence that fitness effects of ectopic exchange are uniformly synergistic has yet to emerge. Furthermore, the intervening 30 years has uncovered small-RNA-mediated silencing as a plausible mechanism for copy-number-dependent transposition [16].

The moment has arrived to reconsider TE dynamics in light of our altered and greatly expanded understanding of transpositional control and purifying selection. In this review, we revisit the transpositional regulation and synergistic selection models, identifying both where theory was prescient and where it should be updated based on recent discovery. We further highlight recent work revealing fascinating and unanticipated connections between small-RNA silencing and purifying selection, which result from heterochromatin formation and heterochromatin spreading of transcriptionally silenced TE loci. Finally, we call for a pluralistic approach to future work, to concurrently evaluate the roles of these forces in shaping the accumulation and distribution of TEs in eukaryotic genomes.

The piRNA pathway

Small-RNA-mediated silencing is a universal strategy for controlling the expression—and ultimately the transposition—of TEs. While pathways and silencing mechanisms differ among organisms, the use of TE insertions themselves as sources of regulatory small RNAs is a common strategy that gives rise to copy-number-dependent repression. The Piwi-interacting RNA (piRNA) pathway is a conserved small-RNA-mediated silencing pathway that regulates TEs in metazoan germlines [reviewed in 16]. TE regulation in the germline is of paramount significance, because only TE insertions produced in gametes can contribute to TE spread in natural populations.

We focus our discussion of piRNA-mediated silencing in Drosophila melanogaster females, where the mechanism is best understood [16]. In D. melanogaster, piRNA-mediated silencing relies on three Piwi-Argonaute proteins: Piwi, Aubergine (Aub) and Argonaute-3 (Ago-3), found in complex with predominantly antisense TE-derived piRNAs. piRNA-Piwi-Argonaute complexes target transcriptional and post-transcriptional silencing through base pairing with nascent and mature TE transcripts, respectively [17–20]. piRNAs are encoded by piRNA clusters, many of which are large and complex loci containing predominantly degraded copies of numerous TE families [17,21]. While there are mechanistic differences, piRNA-mediated TE silencing is present in all sampled metazoans, and is further mirrored by the use of endo-siRNAS to control TEs in plants and budding yeast (Saccharomyces cerevisiae) [22,23].

Interdependence of transcriptional silencing and piRNA transcription

Copy-number dependence in piRNA-mediated TE silencing may arise from the interdependence between transcriptional silencing and the specification of piRNA-producing sites. In both germline and adjacent somatic cells, piRNA-Piwi complexes establish transcriptional silencing of target TEs through multiple changes in chromatin state, including the deposition of the heterochromatic histone mark H3K9me3 [24–29]. However, in germline cells, H3K9me3 serves an additional function in promoting the bidirectional transcription of piRNA precursors, converting the targeted TE into a dual-stranded piRNA cluster [30]. TE insertions are therefore converted from mRNA expressing factories that proliferate TEs, to piRNA producing factories that repress TEs (Fig 2A). Nevertheless, it is unlikely that all TE copies contribute equally to repression, as de novo piRNA cluster formation is proposed to occur only at transcriptionally active TEs, whose nascent transcripts recruit piRNA-Piwi complexes [26,31,32].

Figure 2. Copy-number dependence in transcriptional and post-transcriptional silencing.

A) Euchromatic targets of Piwi-dependent transcriptional silencing are converted into de novo piRNA clusters, thereby increasing the production of homologous piRNAs. Conversion of targets into piRNA clusters is epigenetic, and begins with deposition of H3K9 methylation marks by the histone methyltransferase Eggless. H3K9 marks at targeted loci are then recognized by the HP1 homolog Rhino, which initiates the recruitment of other co-factors. In particular, Moonshiner and Trf2 are paralogs of the eukaryotic transcription factors TAFIIA and TAFIID, respectively, which recruit Pol II for bidirectional transcription of piRNA precursors. B) Sense TE transcripts feed-forward the biogenesis of sense and antisense piRNAs by the ping-pong cycle. Furthermore, antisense precursors in complex with Aubergine are substrates for phasing biogenesis, which increases the abundance of Piwi-bound piRNAs that target transcriptional silencing.

The specific mechanism of interconnectivity between silencing and small-RNA biogenesis in D. melanogaster may not be conserved across eukaryotes, as many of the proteins involved lack homologs even within arthropods [16]. Nevertheless, it illustrates how repeats that are silenced by small RNAs can also act as small RNA source loci, thereby feeding forward their own repression in a dosage-dependent fashion. Indeed, the cyclical relationship between sites of small RNA transcription and sites of transcriptional silencing echoes across the tree of life, and is particularly well-studied in fission yeast (Schizosaccharomyces pombe) and Arabadopsis thaliana [reviewed in 33].

Precursor processing and post-transcriptional silencing

piRNA biogenesis is also connected to TE copy number through the ping-pong cycle, a germline-specific post-transcriptional silencing mechanism that concurrently generates piRNAs from cleaved targets (Figure 2B). Ping-pong is a feed-forward amplification loop, in which Piwi-Argonautes produce piRNAs via reciprocal cleavage of sense and antisense precursors. Because TE-derived mRNAs represent sense precursors, accumulating transcribed TE copies may accelerate ping-pong, enhancing the production of piRNAs and the post-transcriptional silencing of TEs. In D. melanogaster ping-pong may further enhance transcriptional silencing by triggering the production of Piwi-bound piRNAs that target H3K9me deposition in the nucleus (Figure 2B) [34,35]. Ping-pong therefore acts as an adaptive biological sensor that identifies abundant TE transcripts and intensifies their regulation by piRNAs.

While the specific Piwi-Argonautes involved may differ, ping-pong is highly conserved across animals, providing a potentially generalized mechanism for copy-number-dependent TE regulation [16]. Indeed in mice (Mus musculus), post-transcriptional silencing via ping-pong also occurs and is indispensable for long interspersed nuclear element 1 (LINE1) regulation during spermatogenesis [36,37].

Evidence for copy-number-dependent silencing by piRNAs

If piRNA-based TE control is sufficient to limit TE proliferation, silencing must increase with copy number and each new piRNA-generating TE copy must suppress transposition by an increasing amount. There is considerable evidence at least that silencing increases with copy number. In both D. melanogaster and mammals, piRNAs are more abundant from TE families with higher genomic copy numbers and higher expression levels [38,39]. Furthermore, intraspecific variation in copy number has been connected with differences in piRNA-mediated silencing for multiple TE families in Drosophila species. Evidence is particularly strong for P-element DNA transposons and I-element non-LTR retrotransposons in D. melanogaster, whose regulation by maternally transmitted piRNAs increases dosage-dependently with copy number [40,41]. Similarly, for tirant LTR retrotransposons in D. simulans, genomes containing only heterochromatic insertions do not establish piRNA-mediated silencing, while genomes containing additional euchromatic copies are repressive [42]. Finally, in D. virilis, the probability that multiple TE families are repressed is highly dependent on the maternal dose of chromosomes carrying TE copies that are sources of piRNAs (Erwin et al. 2015).

Nonetheless, while piRNA-mediated silencing of some TE families is clearly copy-number dependent (see Box 1), it is not known whether piRNA-mediated TE control satisfies the condition that each new copy drives increasing suppression. Moreover, TE copy number is not the sole determinant of intraspecific variation in piRNA production. For the majority of D. melanogaster TE families there is no observable correlation between copy number and piRNA abundance in comparisons among wild-type strains [43]. What is unique about P, I and tirant elements? These TE families have either recently invaded the genome or are actively being lost, giving rise to dramatic variation in copy number and producing some strains with few or no euchromatic copies [44,45]. Thus, copy-number-dependent piRNA production may predominate at low copy numbers, but other factors, such as the formation of de novo piRNA clusters at some insertions, may be more determinant of differential piRNA production at higher copy numbers [32,43]. Indeed, for P- and I-elements, four copies are sufficient to establish almost complete repression of the sterility effects associated with transposition [41,46].

BOX 1: Copy-number-dependent transposition.

TEs arise in genomes through transposition, and are removed by excision, deletion, or inactivation. The baseline per-copy transposition rate of a TE family (μ) is generally several orders of magnitude greater than the removal rate (ν), allowing for the exponential accumulation of genomic TEs with increasing copy number. However, if the per-copy transposition rate decreases with genomic copy number (n), this provides a potential mathematical solution whereby transposition and removal can be equalized (μ _n = ν), and TEs contained. Motivated by observations of copy-number-dependent transposition in both prokaryotes and D. melanogaster, Charlesworth and Charlesworth [2], and Langley et al [3] proposed functions for copy-number dependent transposition rates that decrease as TEs accumulate. While their mathematical specifics differ, these copy-number-dependent models constrain exponential growth and allow for maintenance of an equilibrium copy number in simulated populations.

The original models were blind to the specific mechanism through which copy-number-dependent transposition was achieved. Since that time, passive copy-number-dependent autoregulation has been documented for DNA transposons, through the accumulation of dominant-negative transposase alleles as TEs proliferate, the saturation of transposase by defective insertions, and the reduced efficiency of transpososome assembly as transposase becomes more abundant [48–51]. However, copy-number-dependent transposition can also result from active repression of transposition, which could be enacted by the host or the TE (although it is important to note that not all proposed forms of repression are copy-number-dependent). In particular this latter context is consistent with small RNA-mediated silencing, in which the production of the small RNA repressors increases with TE copy number [e.g. 38].

It should also be noted that not all small-RNA-mediated mechanisms of TE containment are strictly copy-number-dependent. For example, in maize (Zea mays), a single inverted duplication of the MuDR DNA transposon, designated Mu killer, is sufficient to silence all MuDR elements [47]. Thus, a single TE insertion drastically reduce transposition rates and may be sufficient for TE containment.

Synergistic epistasis and ectopic recombination

Individual TE insertion alleles are typically rare in natural populations, suggesting that purifying selection opposes TE accumulation. Based on their insertion profile, however, individual TE familes are likely to differ in their fitness effects. For example, in D. melanogaster P-elements preferentially insert into 5’-UTRs making them very deleterious [52] whereas Minos elements (from Drosophila hydei) have a preference for introns [53]. In maize, LTR elements show a much greater propensity towards insertion into methylated regions compared to DNA transposons [54]. However, individual differences in fitness costs between insertion alleles is not determinant of whether TEs may be contained at a stable equilibrium copy number. Rather, the combined fitness effects of all TE insertion alleles in the same genome is of critical importance. Specifically, if multiple TE insertions impact fitness independently, this is not generally sufficient restrain TEs in the absence of copy number dependent transposition [2,55]. However, under a form of synergistic epistasis where each new insertion imposes an absolute greater fitness cost compared to the last (a concave fitness function), TE copy numbers are contained over a broad range of transposition rates and deleterious effects [2].

Ectopic recombination among dispersed repetitive sequences (i.e. TE copies of the same family) is the most plausible mechanism for synergistic epistasis among accumulating TEs. While meiotic crossing over between homologous loci offers numerous benefits, non-allelic recombination can result in deleterious structural mutations. For example, deletions that cause Williams-Beuren syndrome, a developmental disorder, result from ectopic recombination between repetitive sequences [56]. Ectopic recombination is prevalent among TEs [9], and the past several decades have yielded significant insight into the mechanisms that control this harmful form of repair. Here we review these mechanisms and highlight evidence that they determine purifying selection at TE insertions. We also critically evaluate whether fitness costs associated with ectopic recombination potentially act synergistically, as is frequently assumed by simulation models of TE population dynamics [11,57].

Heterochromatin and ectopic recombination

The fitness cost of ectopic recombination is underscored by the specialized mechanisms that limit its occurrence. Repetitive centromeric and pericentromeric regions of the genome are known to be refractory to meiotic crossing over. More recently, it has become clear that the chromatin environment can influence two key factors that lower the probability of ectopic recombination: the formation of meiotic Double Strand Breaks (DSBs) and the resolution of DSBs into recombinants.

Spontaneous DSBs occur in all cells, but the germline must also contend with programmed DSBs that initiate crossing over. Organisms employ diverse control mechanisms that limit the induction of programmed DSBs in repetitive sequence; for example they do not occur in heterochromatic regions of the meiotic nucleus in D. melanogaster [58]. Similarly, in fission yeast a distinct cohesin complex in pericentric heterochromatin suppresses DSB formation [59]. Finally, genomic studies in plants, yeast and mammals show that programmed DSBs are depleted in heterochromatic regions containing repeats and TEs [60–62], though there can be heterogeneity among TE classes. For example, in A. thaliana, there is an excess of programmed DSBs in DNA transposons [63].

Ectopic recombination is also reduced by limiting the scope of the homology search for breaks occurring in repetitive DNA (reviewed in [64]). For example, in cell culture, heterochromatic breaks induced by ionizing radiation move to the nuclear periphery, constraining the homology search and the potential for ectopic repair by homologous but non-allelic repeats residing within heterochromatin [65]. Finally, even if non-allelic TEs engage in homologous repair, crossing over may be precluded [66]. In D. melanogaster, while DSBs appear uniformly distributed across euchromatin, the centromere effect blocks their resolution as crossovers in regions near the boundary of euchromatin and heterochromatin [67]. This suppression is a property of the centromere, not the adjacent heterochromatin [68], and is attributed to the action of Bloom’s DNA helicase, a crossover regulator conserved across eukaryotes [69]. Thus, chromatin state and genomic position may lower the probability of ectopic exchange between non-allelic TEs even if they engage in homologous repair. However, little is known about how the chromatin state of TEs dispersed through euchromatin influences their participation in this damaging process.

Evidence for ectopic recombination as a force limiting TE copy number

Reduced ectopic recombination in heterochromatic regions predicts that selection acts more strongly on euchromatic TEs. Therefore, if ectopic recombination provides a critical constraint on TE copy number, it should promote the differential accumulation of TEs in euchromatic and heterochromatic compartments. This frequently appears to be the case [13,70,71]. TEs also segregate at higher frequency in regions of the genome with low recombination rates [14,15], though low gene density and reduced efficacy of natural selection due to the Hill-Robertson effect may contribute to these patterns [11,72] The ectopic recombination model also predicts that, since longer TEs are more likely to experience DSBs and engage in ectopic recombination, they will segregate at lower frequencies. Individual insertions of higher copy number TE families with should as well, due to the higher frequency of non-allelic repair templates. These patterns have been observed in both D. melanogaster [15] and humans [12]. Another test takes advantage of the fact that non-allelic recombination is less likely when insertions are homozygous. Because homozygosity is increased in selfing species, it should reduce the risk of ectopic recombination and allow TE insertions to accumulate [73]. This pattern is observed in the selfing species A. thaliana, which exhibits only a weak relationship between recombination rate and TE density [13]. Moreover, TE frequencies in Arabidopsis lyrata populations that vary in the degree of selfing are higher in selfing populations, consistent with ectopic recombination playing a role in limiting TE insertion frequencies [74]. Other factors, though, must also contribute to the differential accumulation of TEs between these species, as A. thaliana has an overall lower abundance of TEs compared to A. lyrata [75].

While evidence for purifying selection against ectopic recombination seems clear, no study has directly tested if ectopic recombination imposes the synergistic fitness effects required to constrain TE copy number in the absence of other forces. Particularly as TE copies accumulate, the availability of homologous repair templates may no longer be limiting, such that ectopic recombination increases only linearly with copy number (Box 2). Furthermore, meiotic ectopic recombination is likely reduced by the aforementioned mechanisms that constrain DSB formation and ectopic exchange in repetitive regions.

Box 2: Synergistic epistasis and ectopic recombination.

In addition to their analysis of copy-number-dependent transposition, Charlesworth and Charlesworth showed how purifying selection can limit TE copy number. An equilibrium can be obtained when log-fitness declines more rapidly than linearly with copy number [55]. This will be satisfied when TEs have a synergistic effect whereby each new TE copy is more harmful than the previous. Ectopic recombination is proposed to provide such a synergistic fitness function [76] in which the chance of producing a deleterious ectopic recombinant increases with the square of copy number. Is this relationship plausible? In meiosis, the number of double-strand breaks (DSBs) and a minimal rate of DSB repair via alternative pathways may limit ectopic exchange.

Considering the repair of a DSB within a TE, we propose that there is a baseline probability of repair through alternative non-ectopic pathways (ner), such as non-homologous end-joining. Implicitly, when ectopic recombination does not occur, alternative repair occurs instead. This will limit the degree to which increasing copy number leads to an ever increasing rate of ectopic recombination, because at some point homologous TEs will become so abundant in the genome that their availability will no longer be limiting to ectopic repair. We also assume the probability of ectopic recombination from a given DSB does not depend on length or distance of dispersed TE copies, although these assumptions may be open to question [64]. Under these assumptions, a threshold number of TE copies (n_threshhold) would establish a baseline rate of non-ectopic repair, for which additional copies would not increase the chance of ectopic exchange (Figure 1.A). For example, ectopic exchange initiated by a DSB within a TE may be equally likely if there are 100 or 101 genomic TE insertions, with a fixed probability of one minus the probability of non-ectopic repair (1-ner) (Figure 1.A). By contrast, when copy numbers are below the threshold, for a focal DSB within a TE, the probability of ectopic recombination increases linearly with TE copy number (n) (Figure 1.A). As can be seen in Figure 1.B, this model produces synergistic fitness effects only when copy numbers are below the threshold. Above this threshold, fitness declines linearly with copy number. With a linear fitness function, the conditions for limiting TE proliferation become more restrictive [2].

Box 1, Figure I. Meiotic double-strand break number and homologous repair may limit the extent of synergistic epistasis via ectopic recombination.

A) An alternate model of ectopic recombination probability. Here, the baseline probability for repair with no ectopic recombination (ner) is equal to 0.05. The probability for ectopic recombination from a focal TE DSB increases linearly with copy number, until the threshold copy number is met, after which probability of ectopic recombination is constrained by alternative repair pathways to be 1- ner. B) A limit on the extent of ectopic homologous repair may lead to a fitness function that does not universally exhibit synergistic fitness effects. In this example, we assume a genome size of 120 million base pairs, TE insertions 5000 bp in length, 20 meiotic DSBs, a minimum probability of non-ectopic repair (ner) equal to 0.05 and a threshold copy number of 30, where the maximum probability of ectopic recombination is achieved for a given DSB within a TE

At the corner of silencing and selection: the impact of transcriptional silencing on ectopic recombination and gene regulation

TE researchers have typically considered copy-number-dependent repression and synergistic epistasis as separate forces shaping the landscape of TEs in host genomes. However, it is apparent that there is significant interplay between transpositional regulation and purifying selection, which arises from the silenced chromatin state at TE loci.

Transcriptional silencing may reduce ectopic recombination

H3K9me3, the repressive histone mark recruited by Piwi-Argonaute complexes to silenced euchromatic TEs in both D. melanogaster [24–27] and mice [77], is also characteristic of constitutive heterochromatin [78]. Indeed, it was recently demonstrated that transcriptionally silenced euchromatic TEs are more likely to associate with pericentromeric heterochromatin in D. melanogaster nuclei [79]. Therefore, transcriptional silencing of TEs located in regions of high meiotic recombination potentially also reduces the rate of ectopic recombination by subjecting them to robust mechanisms that reduce recombination in heterochromatin. Indeed, in D. melanogaster and fission yeast, mutations impacting small-RNA-mediated silencing pathways and associated H3K9 methyltransferases increase recombination in pericentromeric heterochromatin [80–82]. Similarly, in both A. thaliana and mice the loss of DNA methylation—which is also recruited by piRNAs and siRNAs [16,33]—is associated with a redistribution of crossover events into repeats and TEs [62,83,84]. Therefore, small-RNA-mediated transcriptional silencing of individual TE insertions might decrease their deleterious effects not only by suppressing transposition, but also by suppressing ectopic recombination.

Impacts of transcriptional silencing on host gene expression

Transcriptional silencing may also impact the fitness effects of TEs by altering the expression of adjacent host genes directly by providing or disrupting cis-regulatory elements, or indirectly through transcriptional interference [85]; we refer to these as ‘genic effects’ because they are mediated by the actual TE structure that has inserted, separate from host TE control mechanisms. However, because TEs can nucleate heterochromatin formation in otherwise euchromatic regions, they can also modulate the expression of host genes through small-RNA-mediated ‘epigenetic effects’. These epigenetic effects would furthermore be dependent on the presence and robustness of small RNA-mediated silencing.

Alleles whose expression depends on the silencing of an adjacent TE were first documented in D. melanogaster >30 years ago, using strains that differ in the presence of what is now known to be maternally transmitted piRNAs [86,87]. More recently, population genomic analyses in D. melanogaster, A. thaliana, mice and maize revealed that alleles with adjacent TE insertions exhibit increases in the silencing marks methylation of H3K9, methylation of DNA, or both [88–91]. Furthermore, in D. melanogaster, the spread of these silencing marks is connected to the spread of piRNA production from euchromatic TE insertions into flanking sequences [32].

Altered epigenetic states at and adjacent to TEs are typically viewed as potentially decreasing host gene expression through heterochromatin spreading. But they could also increase host gene expression by reducing the impact of a TE’s genic effect on transcription through transcriptional interference [e.g. 87]. While a priori one might expect more epigenetic effects from spreading because it could have a longer range, the causality of epigenetic effects in reducing host gene expression remains murky. For example, in D. melanogaster, alleles with adjacent TE insertions exhibit lower expression levels in adults when compared to those without TEs [92,93]. Similarly, genes adjacent to methylated TEs in A. thaliana exhibit lower expression levels than those adjacent to under-methylated TEs [94]. However, in both cases, the epigenetic effects of silencing cannot be disentangled from the genic effects of the TE insertion. Furthermore, concurrent comparisons of TE insertion polymorphisms and adjacent gene expression in A. thaliana revealed that TE insertions are equally likely to be associated with increased or decreased expression of flanking genes [88]. Similarly, in a comparison of two wild-type D. melanogaster strains that differ in the positions of their TE insertions, TE insertion alleles with increased H3K9me2 were significantly associated with decreased expression of adjacent genes for only one strain [89]. It seems clear therefore, that the direction, degree, and context in which gene expression is impacted by transcriptional silencing of TEs remains an important direction for future study (Box 3).

Box 3. Isolating epigenetic effects of TE insertions on host gene expression.

To disentangle genic and epigenetic effects for a sample of TE insertions, we took advantage of the P-element hybrid dysgenesis system, in which genetically identical female offspring are produced from reciprocal crosses that differ only in the presence or absence of piRNA-mediated silencing of P-elements [reviewed in 95]. When P strain females are crossed to M strain males, P-elements are silenced in the F1 germline by maternally-transmitted piRNAs. However, in the reciprocal cross, piRNA-mediated silencing is not established due to the absence of P-elements in M strains, resulting in hybrid dysgenesis.

We harnessed published RNA-seq data from non-dysgenic (P x M F1) and dysgenic (M X P F1) ovaries to compare the expression of genes adjacent to P-elements to those not adjacent to P-elements for 128 annotated insertions [96]. This comparison isolates the epigenetic effects of each P-element, because both classes of genetically identical offspring inherit the same P-element insertions in the same genomic positions. If epigenetic effects tend to reduce the expression of adjacent host genes, then those genes should exhibit reduced expression in non-dysgenic crosses where P-elements are silenced by maternally provisioned piRNA. However, we observed the opposite, with genes adjacent to P-elements exhibiting increased, not decreased, expression when P-elements are silenced (Figure 1.A). Epigenetic effects of P-elements may therefore facilitate host gene expression by reducing transcriptional interference from adjacent insertions. Indeed a suite of P-element insertion alleles at the vestigial locus exhibit piRNA dependency that likely reflects transcriptional interference that is repressed by piRNA-silencing [87].

Box 3, Figure I. Genes adjacent to P-elements are differentially expressed in the presence and absence of maternally-deposited piRNA.

A) Differential expression in ovarian tissue between dysgenic (no regulatory piRNA) and non-dysgenic (regulatory piRNA) crosses for genes adjacent to P-elements as compared to genes without an adjacent P-element insertion. Four different window sizes are considered. B) For the 5 Kb window, differential expression is compared between genes with P-element insertions in differing orientation with respect to their transcripts. Note that in both panels, the Y-axis is truncated to 1.5, in order to best capture the expression differences dependent on adjacent P-elements. This excludes some log2FC values for genes not adjacent to P-elements. Significant differences in log2FC values between groups of genes were assessed by permutation tests. *** denotes P < 0.001, ** denotes P < 0.01, N.S. indicates comparison is not significant.

To further isolate the impact of epigenetic effects, we considered whether P-elements were upstream, downstream or internal to host genes, and on the same (+) or opposite (−) strand (Figure 1.B). While reduced gene expression resulting from heterochromatin spreading is predicted to be most pronounced for upstream insertions regardless of orientation, transcriptional interference is most severe in tandem (upstream or internal, +) and convergent (downstream, − ) orientations, and least severe when transcription is divergent (upsteam, −) [97,98]. Consistent with suppressed transcriptional interference, we observed a significantly greater increase in expression among genes with a tandem versus divergent P-element upstream (Figure 1B).

P-elements exhibit an insertion preference upstream of genes [99], and also have a high propensity for dimeric insertions [100]. TE families with distinct insertion preferences could therefore be markedly different in their epigenetic effects. Nevertheless, our analysis emphasizes the importance of disentangling genic and epigenetic effects, and highlights the potential for transcriptional TE silencing to facilitate host gene expression in some contexts.

Epigenetic effects as targets of selection

Transcriptional silencing has potentially opposing effects on the magnitude of purifying selection against euchromatic TEs. On the one hand, it may reduce their fitness cost, not only by decreasing their individual transposition rates, but also by decreasing their participation in ectopic recombination and transcriptional interference of host genes. On the other hand, transcriptional silencing may enhance purifying selection against TE insertions if it reduces host gene expression through heterochromatin spreading. Interestingly, if increased small RNA production leads to ever increasing silencing of adjacent genes, transcriptional silencing may provide an alternative to ectopic recombination as a mechanism for synergistic epistasis [89,93].

This assumption of synergism has yet to be tested.

Consistent with experiencing purifying selection, methylated TE insertions in A. thaliana and H3K9-methylated TEs in D. melanogaster exhibit lower polymorphic frequencies than TEs without epigenetic effects [93,94]. However, it is difficult to disentangle cause and effect in these observations: are epigenetic effects themselves targets of negative selection, or are TE insertions that are more deleterious for other reasons also more likely to nucleate heterochromatin formation? For example, full-length TE insertions are more likely to participate in ectopic recombination, and may also be more likely to cause transcriptional interference on host genes and to recruit heterochromatin formation through nascent transcripts [93].

Concluding remarks

When TEs were first discovered, their maintenance in eukaryotic genomes posed a conundrum. How could they persist without driving the species to extinction? The first theoretical attempts to tackle this problem focused on selection and repression as separate forces, thereby revealing the necessary role of dosage-dependent effects of TE copy number on TE containment. Mathematical relationships between TE copy number and repression or selection were not fully rooted in the underlying mechanisms, but did allow a stable equilibrium of TE copies [2,3]. It has since become clear that TEs are almost universally repressed by small-RNA-mediated silencing, and that a relationship between TE copy number and transposition rate can be produced through small RNA production. However, the mechanisms and quantitative relationships remain poorly understood. Similarly, negative selection against TEs is both powerful and multifaceted, and includes potentially synergistic effects that arise from their epigenetic effects and potential to participate in ectopic recombination. However, the conditions under which purifying selection is synergistic, and the degree to which this depends on small-RNA-mediated silencing remain unclear. The challenge ahead therefore is not only to enhance our quantitative understanding of both transpositional regulation and purifying selection, but also to determine how they concurrently shape TE content and dynamics, and how biological differences among TEs and hosts may result in different outcomes across the tree of life (see Outstanding Questions).

The copy-number-dependent forces that constrain TEs are highly interconnected (Figure 3). The piRNA pathway regulates transposition through a complex feedback loop that connects piRNA biogenesis, TE transcription, and transcriptional and post-transcriptional silencing (Figure 2). Transcriptional silencing by piRNAs furthermore alters the chromatin state of individual TE insertions, thereby potentially imposing fitness costs associated with reduced host gene expression and altering fitness effects associated with ectopic recombination and genic effects. This cacophony of related processes poses a serious challenge to determining which factors are most important for constraining TEs.

Figure 3. Interrelated determinants of TE containment.

TE abundance, distribution, and dynamics in host genomes are determined by the intersection of four types of factors: TE features (green), host factors (blue), repression of transposition (red) and purifying selection (orange). The connectivity between these factors poses a challenge to identifying those that are most influential in the containment of TEs.

A growing number of studies have attempted to harness genomic data sets to disentangle how different forces of selection or the differential strength of repression shape the accumulation of genomic TEs [15,38,39,89,94]. While invaluable, these studies have also struggled to disentangle causative relationships from their covariates. For example, longer TE insertions segregate at lower population frequencies, suggesting they are more deleterious [15]. However, these observations are plausibly explained by either (or both) their enhanced participation in ectopic recombination [15], or by their enhanced epigenetic effects [93]. Disentangling the two requires identifying presently unknown factors that differentiate the risk of ectopic recombination from the potential for epigenetic effects at individual TE insertions. Fortunately, technological advances in genome engineering may provide opportunities to tease apart these factors. For example, using CRISPR, one may manipulate genomes by incorporating dispersed, but non-functional, repetitive sequences. When combined with controlled fitness assays, this may allow the fitness effects of ectopic recombination to be isolated from transposition and altered gene expression. This also has the potential to resolve whether dispersed repeats can establish a fitness function exhibiting synergistic epistasis. Conversely, transgenic and CRISPR-based studies that enable the coordinated manipulation of entire TE families [101] can potentially reveal how the cis effects of TEs, which carry regulatory sequences and are also the target of small RNAs, vary as a function of increasing TE copy number.

Advances in long-read sequencing may empower the examination of another critical question: does the per-element transposition rate decrease with each accumulating TE copy? If so small RNA silencing would be sufficient for TE constraint. The direct measurement of transposition rate is a holy-grail in the study of TE dynamics, which has remained unattainable even in the Illumina era due to the challenge of detecting and quantifying the TE insertions in pooled samples of gametes [102,103]. Long-read sequencing technologies offer considerable promise to address these challenges, both because longer reads provide more power to detect structural variation [104], and also because library preparation does not require PCR amplification, which is known to produce chimeric fragments that mimic transpositions [103]. By applying this technology to genomes whose TE copy numbers have been modulated with CRISPR, the true copy-number dependence of small RNA silencing and transposition may ultimately be resolved.

Along with empirical studies, stochastic simulation models also offer a promising approach to studying the combined impact of alternate forces on TE containment, by allowing for the manipulation of individual factors in the system and thereby avoiding ambiguity of confounding variables. New models are emerging that attempt to incorporate the known biology of piRNA-mediated silencing, explore the consequences of different fitness costs, or both [40,50,57,105]. Copy-number-dependent functions for piRNA mediated silencing have been developed from empirical data [40], and can be improved as we learn more about the underlying biology of dosage-dependence in the piRNA pathway [e.g. 106]. Furthermore, fitness functions for individual TE-imposed fitness costs can be developed, allowing for their impacts to be distinguished [40].

Finally, we propose that further research into interspecific variation across a breadth of taxonomic groups may reveal the impact of host biology on TE proliferation and accumulation. The effective population size is an important determinant of TE accumulation, but numerous other host factors are also potentially significant. Mating systems, mechanisms of chromatin-based gene regulation, and DNA repair pathways vary greatly across the tree of life, providing opportunities to disentangle the factors that control TE proliferation. For example, TEs have been proposed to accumulate more in mammals because the recombination machinery is less prone to ectopic recombination, leading to relaxed selection on repetitive sequences [107]. The converse pattern has been observed in closely-related Drosophila species, in which the recombination landscape is proposed to have evolved rapidly in response to changes in genomic TE content [108]. One mode of defense against TEs in humans appears to be targeted break formation at TEs, leading to their elimination from the genome by biased gene conversion [109]. Such a mechanism of TE defense may not be permitted in species with a high chance of ectopic recombination. Thus, different mechanisms of genome integrity may lead to differential TE accumulation, which in turn enables distinct strategies of TE repression. By integrating an evolutionary perspective, we may reveal that the ever-shifting interaction between the forces of copy-number-dependent transposition, selection and host biology lies at the heart of what determines variation in TE abundance across both genomes and species.

Outstanding Questions.

• What is the quantitative relationship between TE copy number and transposition rate? Is this relationship explained by the acceleration of the ping-pong cycle by sense TE transcripts? Is this relationship explained by the conversion of transcriptionally active TE copies into de novo piRNA clusters? If both ping-pong biogenesis and de novo piRNA cluster formation give rise to copy-number-dependent transposition, how does their relative importance differ across TE families?

• Are fitness effects associated with ectopic recombination among TEs synergistic with respect to TE copy number? What aspects of the host recombination and repair machinery influence this synergism? Does synergism hold across the full range of known copy number variation?

• Does piRNA-dependent heterochromatin formation reduce the participation of euchromatic TEs in ectopic recombination? If so, through what mechanism is ectopic recombination reduced?

• What factors determine the epigenetic effect of piRNA-mediated transcriptional silencing on host gene expression? What is the quantitative relationship between genomic copy number of a given TE and the epigenetic effects of its individual insertions? Is this relationship synergistic? Why do some genes increase expression while others decrease when adjacent TEs are transcriptional silenced?

• Do the mechanisms that ultimately constrain TE proliferation vary among species? How do host-specific differences in recombination machinery, heterochromatin formation, and mating system influence the accumulation of TEs and the evolution of host TE control?

HIGHLIGHTS.

Transposable elements (TEs) can potentially increase exponentially, requiring limits on ever-increasing TE copy number to prevent host extinction.

Two distinct models for TE containment are copy-number-dependent transposition, where increasing TE abundance leads to a decreasing transposition rate, and synergistic epistasis, where increasing TE copy number leads to increasing selection against each individual copy.

TE silencing via small RNAs provides a conserved mechanism for copy-number dependent regulation of transposition.

“Epigenetic” effects associated with TE silencing may impact synergistic epistasis by limiting the potential of silenced insertions to undergo deleterious ectopic recombination.

Epigenetic effects can also modulate the impact of TEs on flanking host gene expression, in complex ways that are not fully understood.

GLOSSARY

Copy-number-dependent transposition

The per-element transposition rate depends on and generally decreases as a function of genomic TE copy number

Double-strand break (DSB)

A form of DNA damage in which phosphodiester bonds between adjacent nucleotides are broken on both strands

Ectopic recombination

Recombination between non-allelic sites, usually between homologous copies of the same interspersed repeat

Heterochromatin spreading

The spread of heterochromatic histone marks from a silenced heterochromatic locus. Trail or phased piRNA biogenesis is one possible mechanism of spreading, which recruits heterochromatic factors by expanding the range of a piRNA source locus

Ping-pong cycle

A piRNA maturation process in which sense and antisense piRNAs are produced by reciprocal cleavage of sense and antisense precursor transcripts by Piwi-Argonaute proteins

Piwi-interacting RNAs (piRNAs)

small 23–29nt RNAs that are found in complex with Piwi-clade Argonaute proteins. piRNAs that are antisense to TE transcripts play a conserved role in germline TE regulation

piRNA-mediated silencing

Transcriptional and post-transcriptional silencing of target transcripts that is targeted by piRNAs that are antisense to their target transcripts

Synergistic purifying selection

Epistatic fitness effects between loci that increase or decrease non-linearly with the number of loci involved. Synergistic epistasis between TE copies of the same family is proposed to be negative, with per-element fitness costs increasing with each additional copy in the genome

Transposition-excision balance

An equilibrium copy number that arises when the per-element transposition rate and the per-element excision rate are equal, such that the average genomic copy number does not change through time

Transposition-selection balance

An equilibrium copy number that arises when the per-element transposition rate is equal to the rate at which TE copies are removed from the population by purifying selection, such that the average genomic copy number does not change through time