Tex19.1 Restricts LINE-1 Mobilisation in Mouse Embryonic Stem Cells

Mobilisation of retrotransposons to new genomic locations is a significant driver of mammalian genome evolution. In humans, retrotransposon mobilisation is mediated primarily by proteins encoded by LINE-1 (L1) retrotransposons, which mobilise in pluripotent cells early in development. Here we show that TEX19.1, which is induced by developmentally programmed DNA hypomethylation, can directly interact with the L1-encoded protein L1-ORF1p, stimulate its polyubiquitylation and degradation, and restrict L1 mobilisation. We also show that TEX19.1 likely acts, at least in part, through promoting the activity of the E3 ubiquitin ligase UBR2 towards L1-ORF1p. Moreover, we show that loss of Tex19.1 increases L1-ORF1p levels and mobilisation of L1 reporters in pluripotent mouse embryonic stem cells implying that Tex19.1 prevents new retrotransposition-mediated mutations from arising in the germline genome. These data show that post-translational regulation of L1 retrotransposons plays a key role in maintaining trans-generational genome stability in the epigenetically dynamic developing mammalian germline.


Introduction
Retrotransposons are mobile genetic elements that comprise around 40% of mammalian genomes (Beck et al. 2011;Hancks and Kazazian 2016;Richardson et al. 2014a). Retrotransposons are a source of genetic variation that shape genome evolution and mammalian development, but their mobilisation can also cause mutations associated with a variety of genetic diseases and cancers (Beck et al. 2011;Hancks and Kazazian 2016;Richardson et al. 2014a;Garcia-Perez et al. 2016).
New retrotransposition events are estimated to occur in around 1 in every 20 human births, and represent around 1% of genetic disease-causing mutations in humans (Kazazian 1999;Hancks and Kazazian 2016). Retrotransposons are classified into three major types depending on their genomic structure: LINEs (long interspersed nuclear elements), SINEs (short interspersed nuclear elements) and LTR (long terminal repeat) retrotransposons. In humans, all new retrotransposition events are catalysed by LINE-1 (L1) elements. Active L1s encode two proteins required for retrotransposition: ORF1p is an RNA binding protein with nucleic acid chaperone activity, and ORF2p is a multidomain protein with reverse transcriptase and endonuclease activities (Beck et al. 2011;Hancks and Kazazian 2016;Richardson et al. 2014a). Both these proteins interact directly or indirectly with various cellular factors and are incorporated into ribonucleoprotein particles (RNPs) along with L1 RNA (Beck et al. 2011;Goodier et al. 2013;Hancks and Kazazian 2016;Richardson et al. 2014a;Taylor et al. 2013). While these proteins exhibit a cis-preference to bind to and catalyse mobilisation of their encoding mRNA, they can act in trans on other RNAs, including those encoded by SINEs (Kulpa and Moran 2006;Wei et al. 2001;Dewannieux et al. 2003). Human L1 also encodes a trans-acting protein, ORF0, that stimulates retrotransposition, although its mechanism of action is currently poorly understood (Denli et al. 2015). Regulating the activity of these L1-encoded proteins will impact on the stability of mammalian genomes and the incidence of genetic disease.
Regulating retrotransposon activity is particularly important in the germline as de novo retrotransposon integrations that arise in these cells can be transmitted to the next generation (Crichton et al. 2014). The mammalian germline encompasses lineage-restricted germ cells including primordial germ cells, oocytes, and sperm, and their pluripotent precursors in early embryos (Ollinger et al. 2010). L1 mobilisation may be more prevalent in pluripotent cells in preimplantation embryos rather than in lineage-restricted germ cells (Kano et al. 2009), and regulation of L1 activity in the pluripotent phase of the germline cycle is therefore likely to have a significant effect on trans-generational genome stability. Repressive histone modifications and DNA methylation typically suppress transcription of retrotransposons in somatic mammalian cells (Beck et al. 2011;Hancks and Kazazian 2016;Richardson et al. 2014a;Crichton et al. 2014), but many of these transcriptionally repressive marks are globally removed during pre-implantation development and during fetal germ cell development in mice (Hajkova et al. 2008;Popp et al. 2010;Santos et al. 2002;Fadloun et al. 2013). DNA methylation in particular plays a key role in transcriptionally repressing L1 in the germline (Bourc'his and Bestor 2004), and it is not clear how L1 activity is controlled in pluripotent cells and fetal germ cells while they are DNA hypomethylated.
In fetal germ cells, loss of DNA methylation correlates with relaxed transcriptional suppression of retrotransposons (Molaro et al. 2014), but also induces expression of methylation-sensitive germline Indeed, the level of L1 expression at different stages of the germline cycle does not completely correlate with the ability of L1 to mobilise and post-translational control mechanisms have been proposed to restrict the ability of L1 to mobilise in the mouse germline (Kano et al. 2009).
However, the molecular identities of these post-translational L1 restriction mechanisms have not yet been elucidated.
Here we show that one of the genes induced in response to programmed DNA hypomethylation in the mouse germline, Tex19.1, regulates L1-ORF1p levels and mobilisation of L1 reporters. We show that mouse TEX19.1, and its human ortholog, physically interact with L1-ORF1p, and regulate L1-ORF1p abundance through stimulating its polyubiquitylation and proteasomedependent degradation. We show that TEX19.1 likely controls L1-ORF1p abundance in concert with UBR2, an E3 ubiquitin ligase that we show also physically interacts with and regulates L1-ORF1p levels in vivo. As anticipated from our analysis, we show that loss of Tex19.1 results in increased L1-ORF1p abundance and increased mobilisation of L1 reporters in pluripotent mouse embryonic stem cells, suggesting that Tex19.1 functions as a post-translational control mechanism to restrict L1 mobilisation in the developing germline.

Germ Cells
We have previously shown that programmed DNA hypomethylation in the developing mouse germline induces expression of a group of genes that are involved in suppressing retrotransposon activity (Hackett et al. 2012). One of the retrotransposon defence genes induced in response to programmed DNA hypomethylation, Tex19.1, suppresses specific retrotransposon transcripts in spermatocytes (Öllinger et al. 2008;Reichmann et al. 2012), however its direct mechanism of action remains unclear. Tex19.1 is expressed in germ cells, pluripotent cells and the placenta and is one of two TEX19 orthologs generated by a rodent-specific gene duplication (Kuntz et al. 2008;Öllinger et al. 2008). These mammal-specific proteins have no functionally characterised protein motifs or reported biochemical activity, but mouse TEX19.1 is predominantly cytoplasmic in the germline (Öllinger et al. 2008;Yang et al. 2010). We therefore investigated if Tex19.1 has posttranscriptional effects on cytoplasmic stages of the retrotransposon life cycle. Since Tex19.1 -/spermatocytes have defects in meiosis that induce spermatocyte death (Öllinger et al. 2008), we analysed mouse L1 ORF1p (mL1-ORF1p) expression in prepubertal testes during the first wave of spermatogenesis before any increased spermatocyte death is evident (Öllinger et al. 2008). Western blotting showed that P16 Tex19.1 -/testes have elevated levels of mL1-ORF1p ( Figure 1A), even though L1 RNA levels do not change ( Figure 1B) (Öllinger et al. 2008;Reichmann et al. 2012) suggesting that Tex19.1 negatively regulates mL1-ORF1p post-transcriptionally in male germ cells.

TEX19.1 Interacts With Multiple Components Of The Ubiquitin-Proteasome System
Post-transcriptional control of protein abundance can occur through regulation of mRNA translation or protein stability. To investigate whether TEX19.1 might be involved in one of these processes we attempted to identify RNAs or proteins that interact with TEX19.1. In contrast to the PIWI proteins MILI and MIWI (Grivna et al. 2006;Unhavaithaya et al. 2009), oligo(dT) pull-downs from mouse testicular lysate suggest that TEX19.1 is not physically associated with RNA in this tissue TEX19.1 protein stability in mouse testes (Yang et al. 2010) which, in combination with the cofractionation and stoichiometric abundance of these proteins in the ESC IPs, suggests that any TEX19.1 protein not associated with UBR2 may be unstable and degraded. TEX19.1-YFP also co-IPs with additional components of the ubiquitin-proteasome system including UBE2A/B, an E2 ubiquitin-conjugating enzyme and cognate partner of UBR2 (Kwon et al. 2003), and a HECTdomain E3 ubiquitin ligase, HUWE1 ( Figure 2B, Supplementary Table 1). The physical associations between TEX19.1 and multiple components of the ubiquitin-proteasome system suggest that the post-transcriptional increase in mL1-ORF1p abundance in Tex19.1 -/testes might reflect a role for TEX19.1 in regulating degradation of mL1-ORF1p.

TEX19.1 Orthologs Directly Interact With L1-ORF1p
We next tested if TEX19.1 might also interact with mL1-ORF1p. Although we did not identify any mL1-ORF1p peptides in the mass spectrometry analysis of TEX19.1-YFP IPs from ESCs, we did identify a single hL1-ORF1p peptide in similar IPs from stable TEX19.1-YFP expressing HEK293T cells. Interactions between E3 ubiquitin ligases and their substrates are expected to be transient and weakly represented in IP experiments, therefore we tested directly whether TEX19.1-YFP and epitope-tagged mL1-ORF1p interact by co-expressing these proteins in HEK293T cells and immunoprecipitating either TEX19.1-YFP or epitope-tagged mL1-ORF1p. Both IPs revealed weak reciprocal interactions between TEX19.1-YFP and epitope-tagged mL1-ORF1p ( Figure 2D, Supplementary Figure S2C). Although human TEX19 is significantly truncated relative to its mouse ortholog, the physical interaction between TEX19 and L1-ORF1p is conserved in humans (Supplementary Figure S2D, Supplementary Figure S2E). We next tested whether the biochemical interaction between TEX19.1-YFP and mL1-ORF1p-T7 is reflected by co-localisation of these proteins. TEX19.1 is predominantly cytoplasmic in ES cells and in germ cells (Öllinger et al. 2008;Yang et al. 2010), but in the hypomethylated placenta and when expressed in somatic cell lines TEX19.1 can localise to the nucleus (Kuntz et al. 2008;Reichmann et al. 2013). The context-dependent localisation of TEX19.1 suggests that TEX19.1interacting proteins in ES cells and germ cells could retain this protein in the cytoplasm in these cell types. L1-ORF1p has been reported to form cytoplasmic aggregates that co-localise with stress granule markers (Doucet et al. 2010), therefore we tested whether co-expression of L1-ORF1p and TEX19.1 might localise TEX19.1 to these L1-ORF1p-containing aggregates. As expected, confocal microscopy showed that TEX19.1-YFP localises to the nucleus when expressed in U2OS cells, however co-expression with mL1-ORF1p-T7 resulted in partial co-localisation of both these proteins in cytoplasmic aggregates in around 40% of cells ( Figure 2F). In more extreme examples, co-expression of mL1-ORF1p-T7 re-localised all detectable TEX19.1-YFP out of the nucleus and into cytoplasmic aggregates ( Figure 2F). These co-localisation data are consistent with the co-IP data suggesting that TEX19.1-YFP and mL1-ORF1p-T7 physically interact.
A number of host factors have been shown to associate with L1-ORF1p, although many of these interactions are indirect and mediated by RNA, likely reflecting interactions within the L1 RNP (Goodier et al. 2013;Taylor et al. 2013). However, TEX19.1-YFP also interacts with a mutant allele of mL1-ORF1p which has severely impaired binding to RNA and impaired mobilisation (Kulpa and Moran 2005;Martin et al. 2005) (Figure 2E, Supplementary Figure S2F), suggesting that the interaction between TEX19.1-YFP and mL1-ORF1p is RNA-independent and could potentially be direct. We therefore tested whether bacterially expressed TEX19 might interact with bacterially expressed hL1-ORF1p. Co-expression of double-tagged human MBP-TEX19-GB1-His 6 with Streptagged human L1-ORF1p (Strep-hL1-ORF1p) in bacteria resulted in a strong interaction between -9 -175 180 185 190 these proteins, and isolation of a stable TEX19-hL1-ORF1p complex ( Figure 2G). Taken together, the co-IPs, co-localisation and isolation of a TEX19-L1-ORF1p from bacterially expressed proteins suggest that TEX19 directly interacts with L1-ORF1p and, to our knowledge, represents the first example of a mammalian host protein that directly binds to a protein encoded by L1 retrotransposons.

Tex19.1 Orthologs Stimulate Polyubiquitylation and Degradation of L1-ORF1p
The strong interaction between TEX19 and hL1-ORF1p seen with bacterially-expressed proteins contrasts with weaker interactions detected in HEK293T cells. However, it is possible that the difference in the strength of these interactions reflects the presence of UBR2 in HEK293T cells, which allows a TEX19-UBR2 complex to assemble and transiently interact with hL1-ORF1p to catalyse its ubiquitylation and subsequent degradation. We therefore investigated if L1-ORF1p is ubiquitylated and degraded by the proteasome, and whether this might be stimulated by TEX19.
Endogenously expressed mL1-ORF1p in mouse testes represents a collection of protein molecules expressed from hundreds of variant copies of L1 at different genomic loci, therefore to allow us to correlate the abundance of L1-ORF1p with its encoding RNA more accurately, and to detect transient polyubiquitylated intermediates that are destined for proteasome-dependent degradation, we expressed engineered epitope-tagged hL1-ORF1p constructs in HEK293T cells. HEK293T cells do not endogenously express detectable levels of TEX19 (Reichmann et al. 2017) and in vivo ubiquitylation assays show that there is basal ubiquitylation of hL1-ORF1p, detectable as a ladder of hL1-ORF1p species in his 6 -myc-Ub pull-downs, in these cells ( Figure 3A). The increasing molecular weights of these bands presumably correspond to increasing ubiquitylation of hL1-ORF1p. Furthermore, treating these cells with the proteasome inhibitor MG132 showed that hL1-ORF1p abundance is negatively regulated by the proteasome in the absence of TEX19 ( Figure 3B).
Interestingly, co-expression of TEX19 during the in vivo ubiquitylation assay increases polyubiquitylation of hL1-ORF1p ( Figure 3C). TEX19 increases the proportion of hL1-ORF1p-T7 that has at least four ubiquitin monomers, the minimum length of polyubiquitin chain required to target proteins to the proteasome (Thrower et al. 2000). Furthermore, expression of TEX19 in these cells also reduces the abundance of hL1-ORF1p protein without any change in the abundance of its encoding RNA ( Figure 3D). These gain-of-function data for TEX19 mirror the loss-of-function data obtained from Tex19.1 -/testes, confirm that the increased mL1-ORF1p levels in Tex19.1 -/testes are not a consequence of altered progression of Tex19.1 -/spermatocytes through meiosis (Crichton et al. 2017b;Öllinger et al. 2008), and suggest that Tex19.1 orthologs function to post-translationally regulate L1-ORF1p abundance. The ubiquitylation and interaction data together suggests that, TEX19 orthologs regulate L1-ORF1p abundance through binding to L1-ORF1p and stimulating its polyubiquitylation and proteasome-dependent degradation.

Tex19.1 Orthologs Restrict Mobilisation of L1 Reporters
L1-ORF1p has essential roles in L1 retrotransposition (Beck et al. 2011;Richardson et al. 2014a; Hancks and Kazazian 2016), therefore since TEX19 orthologs bind to L1-ORF1p and negatively regulate its abundance, we investigated whether Tex19.1 might inhibit L1 mobilisation in cultured cells. Engineered L1 retrotransposition assays with an EGFP retrotransposition reporter in HEK293T cells ( Figure 4A) were used to measure the effect of Tex19.1 on mobilisation of mouse L1 reporters. Expression of Tex19.1 reduced the ability of both synthetic mouse L1 and a G f type mouse L1 reporter to mobilise in these cells, but had no detectable effect on negative control reporters carrying inactivating mutations in ORF2 ( Figure 4B). Tex19.1 did not affect mobilisation of L1 retrotransposition reporters as potently as the L1 restriction factor APOBEC3A (Bogerd et al. 2006b), but still reduced L1 mobilisation by around 50% in this assay ( Figure 4B). Mouse Tex19.1 also restricts mobilisation of engineered human L1 reporters (Supplementary Figure S4A less efficiently than it restricts mouse L1 reporters. These data show that Tex19.1 can function as a restriction factor for L1 mobilisation. Mouse Tex19.1 expression is activated in response to DNA hypomethylation in multiple contexts (Hackett et al. 2012), and in humans TEX19 is a cancer testis antigen expressed in multiple types of cancer (Feichtinger et al. 2012). We therefore tested whether expression of TEX19 orthologs might be sufficient to restrict L1 mobilisation in multiple host cell types. L1 retrotransposition assays using blasticidin resistance reporters in HeLa cells ( Figure 4C) showed that mouse Tex19.1 similarly restricts mobilisation of a mouse and human L1 reporter by ~50% in this epithelial carcinoma cell line ( Figure 4D). Human TEX19 also restricts mobilisation of mouse and human L1 reporters by ~50% in these cells ( Figure 4D). Similar effects on mobilisation of L1 reporters were also observed in U2OS osteosarcoma cells (Supplementary Figure S4B, Supplementary Figure   S4C). Thus, TEX19 orthologs are host restriction factors for L1 retrotransposition in mice and humans. Importantly, although we have also shown that TEX19 orthologs promote polyubiquitylation and degradation of L1-ORF1p, since TEX19 can directly bind to L1-ORF1p it is possible that this interaction also disrupts aspects L1-ORF1p function and contributes to TEX19dependent restriction of L1 mobilisation. Moreover, there could be additional aspects of TEX19 function that may also be contributing to its ability to restrict L1 mobilisation. Indeed, it is not uncommon for host restriction factors to influence multiple aspects of retrotransposon or retroviral life cycles (Wang et al. 2010;Burdick et al. 2010;Goodier et al. 2012;Holmes et al. 2007). To dissociate the effects of UBR2 on stability of TEX19.1 protein from potential effects on L1-ORF1p abundance and L1 mobilisation we therefore tested whether UBR2 can regulate L1 in the absence of effects on TEX19 stability by using somatic HEK293T cells. Interestingly, mouse UBR2 co-IPs with mL1-ORF1p in HEK293T cells ( Figure 5A), a cell type that does not express any detectable TEX19 (Reichmann et al. 2017), suggesting that UBR2 is able to regulate L1-ORF1p independently of any effects on TEX19 protein stability. UBR2 also interacts with mL1-ORF1 RA p -13 -  Figure S5C). In addition, overexpression of UBR2 alone restricts mobilisation of a human L1 retrotransposition reporter ( Figure 5C). Thus, at least when it is overexpressed, UBR2 is able to physically interact with L1-ORF1p and restrict mobilisation of L1 reporters.
To test whether endogenously expressed UBR2 might regulate hL1-ORF1p abundance we generated UBR2 mutant HEK293T cell lines by CRISPR/Cas9-mediated genome editing. However, these cell lines grew slowly and poorly in culture, presumably reflecting the normal cellular roles of UBR2 in cohesin regulation, DNA repair, and chromosome stability (Ouyang et al. 2006;Reichmann et al. 2017). Therefore, to allow a meaningful analysis of the role of endogenous UBR2 in L1 regulation we analysed Ubr2 -/mice (Supplementary Figure S5D, Supplementary Figure SE) which, despite having defects in spermatogenesis and female lethality, are otherwise grossly normal (Kwon et al. 2003). mL1, but not Tex19.1, is expressed in the brain (Wang et al. 2001;Muotri et al. 2010), therefore we used this tissue to assess whether Ubr2 might have a Tex19.1-independent role in regulating mL1-ORF1p. Consistent with the physical interaction between UBR2 and mL1-ORF1p ( Figure 5A), we found that mL1-ORF1p abundance is post-transcriptionally elevated in the cerebellum of Ubr2 -/mice ( Figure 5D), suggesting that UBR2 may directly regulate polyubiquitylation and subsequent degradation of mL1-ORF1p in vivo. Interestingly, loss of Ubr2 has no detectable effect on mL1-ORF1p abundance in the cerebrum (Supplementary Figure S4E), which may reflect cell type specific differences in L1 regulation or genetic redundancy between UBR-domain proteins (Tasaki et al. 2005). Nevertheless, regardless of this additional complexity in the cerebrum, the increased abundance of mL1-ORF1p in Ubr2 -/cerebellum demonstrates that endogenous Ubr2 plays a Tex19.1-independent role in regulating mL1-ORF1p abundance in vivo. Ubr2 has numerous endogenous cellular substrates and host functions beyond regulating mL1-ORF1p (Ouyang et al. 2006;Reichmann et al. 2017;Sriram et al. 2011), but expression of Tex19.1 in the germline or in response to DNA hypomethylation appears to stimulate a pre-existing activity of UBR2 to regulate mL1-ORF1p, possibly at the expense of UBR2's activity towards some endogenous cellular substrates (Reichmann et al. 2017).

Pluripotent Cells
As outlined earlier, L1 mobilisation is thought to occur primarily in pluripotent cells within the germline cycle (Kano et al. 2009), and regulation of L1 expression and mobilisation in these cells is likely to significantly impact on the ability of L1 to influence germline mutation and genome evolution. Therefore, we tested whether Tex19.1, which is expressed in pluripotent cells (Kuntz et al. 2008), has a role in regulating L1 expression and restricting L1 mobilisation in this cell type. We first investigated whether Tex19.1 regulates mL1-ORF1p abundance in pluripotent mouse ESCs.
Biochemical isolation of polyubiquitylated proteins and treatment with proteasome inhibitor suggests that endogenous mL1-ORF1p is polyubiquitylated and regulated by the proteasome in pluripotent mouse ESCs ( Figure 6A, Figure 6B). hL1-ORF1p abundance is similarly regulated by the proteasome in human ESCs and human embryonal carcinoma cells (Supplementary Figure S6).
In contrast to a previous report assessing the abundance of retrotransposon RNAs in ESCs derived from heterozygous mouse crosses (Tarabay et al. 2013), Tex19.1 -/mouse ESCs generated by sequential gene targeting (Supplementary Figure S7) in a defined genetic background and analysed at low passage number do not de-repress L1 RNA ( Figure 6C). These Tex19.1 -/mouse ESCs contain elevated levels of endogenous mL1-ORF1p, but this increase in mL1-ORF1p levels is not accompanied by increased endogenous L1 mRNA levels ( Figure 6C We next tested whether loss of Tex19.1 also results in increased mobilisation of mouse L1 reporters in pluripotent ESCs. Although L1 retrotransposition assays have previously been performed in pluripotent human cells (Wissing et al. 2011;Garcia-Perez et al. 2007Wang et al. 2014;Klawitter et al. 2016), this assay had not been adapted to mouse ESCs and, to our knowledge, no restriction factor has yet been shown to restrict mobilisation of L1 reporters in mouse pluripotent cells or germ cells. Adaptation of the L1 retrotransposition assay using a blasticidin resistance retrotransposition reporter ( Figure 4C) to mouse ESCs resulted in the appearance of blasticidin resistant colonies that could be suppressed by either introducing the N21A mutation (Alisch et al. 2006) into the endonuclease domain of ORF2, or co-transfection of the L1 restriction factor APOBEC3A (Bogerd et al. 2006b) (Supplementary Figure S9). Thus, the adapted L1 retrotransposition assay appears to reflect bone fide mobilisation of L1 reporters. Interestingly, mobilisation of L1 reporter constructs is elevated around 1.5-fold in Tex19.1 -/-ESCs ( Figure Figure 6D). Thus, these data suggest that the role of endogenously expressed Tex19.1 in mouse pluripotent cells is to restrict L1 mobilisation, and thereby promote trans-generational genome stability.

Discussion
This study identifies Tex19.1 as a host restriction factor for L1 in the mammalian germline. We have previously reported that Tex19.1 plays a role in regulating the abundance of retrotransposon RNAs (Öllinger et al. 2008;Reichmann et al. 2012Reichmann et al. , 2013, which appears to reflect transcriptional derepression of specific retrotransposons (Crichton et al. 2017a). Although loss of Tex19.1 results in de-repression of L1 RNA in placenta (Reichmann et al. 2013), L1 RNA abundance is not affected by loss of Tex19.1 in male germ cells (Öllinger et al. 2008) or, in contrast to a previous report (Tarabay et al. 2013), in mouse ESCs ( Figure 6). Indeed here we show that Tex19.1 has a role in post-translational regulation of L1-ORF1p steady-state levels in these cells. Thus, Tex19.1 appears to regulate retrotransposons at multiple stages of their life cycle. It is possible that Tex19.1 is affecting different E3 ubiquitin ligases, or different E3 ubiquitin ligase substrates, in order to repress different stages of the retrotransposon life cycle. However, loss of Tex19.1 results in a 1.5-fold increase in mobilisation of L1 reporters in pluripotent cells. Since L1 mobilisation mostly takes place in the pluripotent phase of the germline cycle, and new L1-dependent mobilisation events are thought to be inherited by one in every twenty human births (Kazazian 1999), TEX19 activity could potentially be preventing new retrotransposition events from being inherited by up to 3 million births annually. Retrotransposons appear to provide functions that are advantageous for mammalian development and evolution (Garcia-Perez et al. 2016), and the activity of restriction mechanisms like the TEX19-dependent mechanism we have described here, that control the ability of retrotransposons to mobilise, rather than eliminate their transcriptional activity altogether, could potentially allow retrotransposons to participate in and drive the evolution of key gene regulatory networks in pluripotent cells while minimising their mutational load on the germline genome.
Our data suggests that L1-ORF1p is post-translationally modified by ubiquitylation in somatic and germline cells. Phosphorylation of L1-ORF1p has been previously reported in somatic tissues and is -17 -365 370 375 380 required for L1 retrotransposition in these cells (Cook et al. 2015). However, we are not aware of any previous reports that post-translational modifications of L1-ORF1p are present in the germline, particularly in the pluripotent phase of the germline cycle when L1 retrotransposition is thought to primarily occur (Kano et al. 2009). Post-translational regulation of L1 potentially provides an additional layer of genome defence that could be particularly important during periods of epigenetic reprogramming in early embryogenesis or in the developing primordial germ cells when transcriptional repression of retrotransposons might be more relaxed (Molaro et al. 2014;Fadloun et al. 2013). Indeed, the sensitivity of Tex19.1 expression to DNA hypomethylation (Hackett et al. 2012) will allow post-translational suppression of L1 to be enhanced during these stages of development. Post-translational regulation of L1 is also likely important to limit the activity of L1 variants that evolve to escape transcriptional repression and will provide a layer of genome defence while the host adapts its KRAB zinc-finger protein repertoire to these new variants (Jacobs et al.

2014)
. Analysis of L1 evolution shows that regions within L1-ORF1p are under strong positive selection suggesting that host restriction systems are targeting L1-ORF1p post-translationally and impacting on evolution of these elements (Boissinot and Furano 2001;Sookdeo et al. 2013).
Although this evidence for post-translational restriction factors acting on L1-ORF1p has been known for over 15 years, to our knowledge no host factors have been identified that directly bind to L1-ORF1p and restrict L1 mobilisation. It is possible that the physical interactions between L1-ORF1p and TEX19:UBR2 that we describe here are contributing to these selection pressures acting on L1-ORF1p. While UBR2 is able to target L1-ORF1p in the absence of TEX19, evolution of a less constrained TEX19 adapter to provide a further link between UBR2 and L1-ORF1p could potentially resolve the contradictory pressures on UBR2 to maintain interactions with some endogenous cellular substrates while targeting a rapidly evolving retrotransposon protein for degradation. Our data indicates that TEX19.1 likely exists in a complex with UBR2 in ESCs, and that TEX19.1 stimulates a basal activity of UBR2 to bind to and polyubiquitylate L1-ORF1p ( Figure 7). Ubr1, a yeast ortholog of UBR2, has different binding sites for different types of substrate (Xia et al. 2008).
Ubr1 can bind to proteins that have specific residues at their N-termini (N-end rule degrons), and to proteins that have more poorly defined non-N-terminal internal degrons. Full-length human L1-ORF1p does not have a potential N-end rule degron at its N-terminus (Kim et al. 2014), and the interaction between UBR2 and L1-ORF1p likely reflects an internal degron in the retrotransposon protein. One of the known internal degron substrates of yeast Ubr1 is CUP9, a transcription factor that regulates expression of a peptide transporter (Turner et al. 2000). Binding of Ubr1 to the CUP9 internal degron is activated by specific dipeptides binding to the N-end rule degron binding sites, which inhibits binding of N-end rule degrons, allosterically relieves the activity of an autoinhibitory domain within Ubr1, stimulates binding to the CUP9 internal degron, and increase processivity of Ubr1 resulting in increased polyubiquitylation of CUP9 (Du et al. 2002;Xia et al. 2008;Turner et al. 2000). This generates a positive feedback loop to activate transcription of a peptide transporter when these peptides are present. The effect of these dipeptides on Ubr1 activity in yeast strongly resonates with the effects of TEX19 orthologs on UBR2 activity in mammals. TEX19 binds to UBR2 and inhibits its activity towards N-end rule substrates (Reichmann et al. 2017), but stimulates polyubiquitylation of L1-ORF1p, possibly through an internal degron in L1-ORF1p. The direct interaction between TEX19.1 and L1-ORF1p could further enhance L1-ORF1p binding to UBR2 by stabilising the highly flexible L1-ORF1p trimers (Khazina et al. 2011) in a conformational state that exposes an internal degron and favours their ubiquitylation. Thus, TEX19 orthologs appear to function, at least in part, by re-targeting UBR2 away from N-end rule substrates and towards a retrotransposon substrate. However, the direct interaction between TEX19.1 and L1-ORF1p means that it is possible that TEX19.1 is interfering with L1-ORF1p function in multiple ways in order to restrict L1 mobilisation. Thus, while one outcome of this interaction appears to be increased -19 -polyubiquitylation and degradation of L1-ORF1p, the interaction between TEX19.1 and L1-ORF1p could also interfere with the nucleic acid chaperone activity of L1-ORF1p (Martin et al. 2005), or its interactions with either L1-encoded or host-encoded molecules. Further work is needed to determine whether these or other additional mechanisms are contributing to the ability of TEX19.1 to restrict L1 mobilisation.
The data presented here suggests that programmed DNA hypomethylation in the mouse germline extends beyond activating components of the PIWI-piRNA pathway (Hackett et al. 2012) to include enhancing the activity of the ubiquitin-proteasome system towards retrotransposon substrates.
Activation of post-translational genome-defence mechanisms may allow mammalian germ cells to safely transcribe retrotransposons by preventing these transcripts from generating RNPs that can mutate the germline genome (Supplementary Figure S10). The retrotransposon transcripts can then be processed into piRNAs and used to identify retrotransposon loci where epigenetic silencing needs to be established. De novo establishment of epigenetic silencing at retrotransposons in the Arabidopsis germline involves transfer of small RNAs between a hypomethylated vegetative cell and a germ cell (Slotkin et al. 2009), whereas these processes happen sequentially in the same germ cell in mammals (Supplementary Figure S10). Therefore the ability to enhance post-translational control of retrotransposons may be a key feature of epigenetic reprogramming in the mammalian germline that limits the trans-generational genomic instability caused by retrotransposon mobilisation.

Acknowledgments
We thank the MRC (intramural programme grants to I.R.A (MC_PC_U127580973) and to R.R.M.

Mice
TEX19.1 mutant mice on a C57BL/6 genetic background were maintained and genotyped as described (Öllinger et al. 2008;Reichmann et al. 2012). Tex19.1 +/heterozygotes have no detectable testis phenotype and indistinguishable sperm counts from wild-type animals (Öllinger et al. 2008), and prepubertal Tex19.1 -/homozygotes were typically compared with heterozygous littermates to control for variation between litters. Ubr2 -/mice were generated by CRISPR/Cas9 double nickasemediated genome editing in zygotes (Ran et al. 2013 Table 2) were microinjected into the cytoplasm of C57BL/6 × CBA F2 zygotes. The repair template introduces an XbaI restriction site and mutates cysteine-121 within the UBR domain of UBR2 (Uniprot Q6WKZ8-1) to a premature stop codon. The zygotes were then cultured overnight in KSOM (Millipore) and transferred into the oviduct of pseudopregnant recipient females. Pups were genotyped for the presence of the XbaI restriction site. The Ubr2 -/male mice generated in this way have no overt phenotypes except testis defects and infertility and Ubr2 -/females are born at sub-Mendelian ratios, all as previously described for Ubr2 -/mice generated by gene targeting in ESCs (Kwon et al. 2003

Cell Culture
We used cell lines that were previously shown to support retrotransposition of engineered L1 constructs or Tex19.1 -/models generated in this study. Cell lines were maintained at 37°C in 5% Hamster XR-1 cells (Stamato et al. 1983) were provided by Thomas D. Stamato (The Lankenau Institute fro Medical Research, US) and grown in DMEM low glucose medium containing 10% foetal calf serum, 1% L-glutamine, 1% penicillin-streptomycin and 0.1 mM non-essential amino acids. Human PA-1 cells (Zeuthen et al. 1980) were obtained from ATCC and grown in Minimal Essential Media (MEM) supplemented with 10% heat-inactivated foetal calf serum, 1% Lglutamine, 1% penicillin-streptomycin and 0.1 mM non-essential amino acids. H9 human ESCs (Thomson et al. 1998) were obtained from Wicell and cultured and passaged as previously described using conditional media (CM) (Garcia-Perez et al. 2007). To prepare CM, human foreskin fibroblasts (obtained from ATCC) were mitotically inactivated with 3000-3200 rads γ-irradiation, seeded at 3x10 6 cells/225 cm 2 flask and cultured with hESC media (KnockOut DMEM supplemented with 4 ng/ml bFGF, 20% KnockOut serum replacement, 1 mM L-glutamine, 0.1 mM β-mercaptoethanol and 0.1 mM non-essential amino acids) for at least 24 hours before media harvesting. We collected CM 24, 48 and 72 hours after seeding. H9 human ESCs (Wicell) were maintained on Matrigel (BD Biosciences)-coated plates in human foreskin fibroblast-conditioned -24 -515 520 525 530 media. The absence of Mycoplasma in cultured cells was confirmed once a month using a PCRbased assay (Minerva). Single tandem repeat genotyping was done at least once a year to ensure the identity of the cell lines used.

Generation Of Stable Cell Lines
ES and HEK293 cell lines stably expressing TEX19.1-YFP or YFP alone were generated by transfecting E14 ESCs or HEK293 cells with linearised pCAG-TEX19.1-YFP and pCAG-YFP expression plasmids (Supplementary Table 3) and selecting for the G418 resistance cassette. Stable cell lines were flow sorted to select for YFP expression. For pCAG-YFP transfection, the cell lines were flow sorted to select for cells expressing YFP at similar levels to the pCAG-TEX19.1-YFP cell lines. Stable Flp-In-293 cells (Invitrogen) expressing T7-tagged hL1-ORF1p from a CMV promoter at the Flp-In locus were generated using the pcDNA5⁄FRT Flp-In vector, and selected using 100 μg/ml hygromycin and 100 μg/ml Zeocin according to the supplier's instructions. Validation of these cell lines is shown in Extended Data Figure 10.

qRT-PCR
RNA was isolated from cells or tissues using TRIzol reagent (Life Technologies), treated with DNAse (DNAfree, Ambion) and used to generate random-primed cDNA (First Strand cDNA Kit, Life Technologies) as described by the suppliers. qPCR was performed on the cDNA using the SYBR Green PCR System (Stratagene) and a CFX96 Real-Time PCR Detection System (Bio-Rad).
Control qRT-PCR reactions were performed in the absence of either reverse transcriptase or qPCR template to verify the specificity of any qRT-PCR signals obtained. Primers were validated to perform at >90% efficiency in the qRT-PCR assay, and expression quantified using the 2 -∆∆Ct method (Livak and Schmittgen 2001). Alternatively, qPCR was performed using SYBR Select Master Mix (Applied Biosystems) and a Light Cycler 480 II (Roche), and expression quantified using the relative standard curve method as described by the suppliers.

Immunostaining
Immunostaining on P16 testes was performed by fixing decapsulated P16 testes in 4% paraformaldehyde in PBS, embedding the tissue in paraffin wax, and cutting 6 μm sections on a microtome. Sections were de-waxed in xylene, rehydrated, and antigen retrieval was performed by boiling slides in a microwave for 15 mins in 10 mM sodium citrate pH 6. Sections were blocked in PBS containing 10% goat serum, 3% BSA, 0.1% Tween-20, then incubated in 1:300 rabbit anti-L1 unit area, with slides immunostained with non-specific rabbit IgG and secondary antibodies used to calculate and subtract background.
Oligo(dT)-cellulose beads (Ambion) were blocked in lysis buffer containing 5% BSA for 1 h at 4°C, then incubated with lysate for 1 h at 4°C. Oligo(dT)-cellulose beads were washed three times with lysis buffer, and bound proteins eluted by boiling in Laemmli SDS sample buffer and analysed

Isolation Of TEX19.1-YFP Complexes
Cytoplasmic extracts were prepared as described (Wright et al. 2006). Briefly, stable YFP or TEX19.1-YFP ESCs growin in LIF+serum conditions were resuspended in 3 volumes buffer A (10 mM HEPES pH 7.6, 15 mM KCl, 2 mM MgCl 2 , 0.1 mM EDTA, 1 mM DTT, 0.2 mM PMSF, Complete protease inhibitors (Roche)) and incubated on ice for 30 mins. Cells were lysed in a Dounce homogeniser, one-tenth volume buffer B (50 mM HEPES pH 7.6, 1 M KCl, 30 mM MgCl 2 , 0.1 mM EDTA, 1% NP-40, 1 mM DTT, 0.2 mM PMSF) added, then the lysate centrifuged twice for 15 minutes at 3400g at 4°C to deplete nuclei. Glycerol was added to a final volume of 10%, the extracts centrifuged at 12,000g for 5 minutes at 4°C, pre-cleared with protein A agarose beads (Sigma) then with blocked agarose beads (Chromotek), before incubation with GFP-Trap agarose beads (Chromotek) for 90 minutes at 4°C. Beads were collected by centrifugation at 2700g for 2 minutes at 4°C, washed three times with 9:1 buffer A:buffer B, and protein eluted by boiling in 2× Laemmli SDS sample buffer for 3 minutes. Protein samples were separated on pre-cast Bis-Tris polyacrylamide gels (Invitrogen) and stained with Novex colloidal blue staining kit (Invitrogen).
Lanes were cut into seven regions according to migration of molecular weight markers and in-gel digestion with trypsin, and mass spectrometry using a 4800 MALDI TOF/TOF Analyser (ABSciex) equipped with a Nd:YAG 355nm laser was performed by St. Andrews University Mass Spectrometry and Proteomics Facility. Mass spectrometry data was analysed using the Mascot search engine (Matrix Science) to interrogate the NCBInr database using tolerances of ± 0.2 Da for peptide and fragment masses, allowing for one missed trypsin cleavage, fixed cysteine carbamidomethylation and variable methionine oxidation.

Luciferase Assays
Luciferase activity was measured 24 hours post-transfection using the Dual-Luciferase Reporter Assay system (Promega) following manufacturer's instructions.
In all retrotransposition assays, we also included a transfection efficiency control to calculate rates of engineered retrotransposition as described ( Media was replaced after 8 hours and transfected mouse ESCs passaged into a gelatin-coated 100mm tissue culture plate 24 hours later. 200 μg/ml G418 or 8 μg/ml blasticidin S selection for 12 days was initiated after an additional 24 hours, and drug-resistant foci fixed, stained and counted as described for HeLa cells. -33 -Retrotransposition assays with mEGFPI tagged L1 constructs in cultured HEK293T cells were performed as described (Goodier et al. 2013;Wei et al. 2000). 2×10 5 HEK293T cells were plated in a 35 mm diameter well, then transfected with Lipofectamine 2000 (Invitrogen) and 1μg plasmid DNA per well using OptiMEM (Invitrogen) following the manufacturer instructions 20 hours later. 24 hours later, fresh media containing 2 μg/ml puromycin (Sigma) was added daily for 7 days. Cells were collected by trypsinization and the percentage of EGFP-expressing cells determined using a FACSCanto II flow cytometer (BD Biosciences). Transfection with mutant L1 plasmid (99-gfp-JM111 or 99-gp-L1SMmut2) allowed a threshold to be established for background fluorescence.

Supplementary Table 3. Plasmids Used In This Study.
Plasmid Description

pCEPL1SM-T7
A derivative of pCEPL1SM where a T7-epitope tag has been cloned in the Cterminus of ORF1p. Also, the L1 element lacks the 5' UTR from L1spa.
pCEPL1SM-T7-ORF1 RA A derivative of pCEPL1SM-T7 that contains two missense mutations in the RNA binding domain of ORF1p (R297A, R298A). These mutations abolish the ability of the codon-optimised L1 TF element to mobilise.

p3xFLAG-CMV-Ubr2
Mouse Ubr2 coding sequence cloned in to p3XFLAG-CMV-10 (Sigma) expression vector that generate a 3xFLAG epitope tag at the N-terminus of UBR2.

pEGFP3N1-Ubr2
Mouse Ubr2 coding sequence cloned into pEGFP-N1 (Clontech) expression vector to generate a GFP tag at the C-terminus of UBR2.