Dissection of purified LINE-1 reveals distinct nuclear and cytoplasmic intermediates

Long Interspersed Nuclear Element-1 (LINE-1, L1) is a mobile genetic element active in human genomes. L1-encoded ORF1 and ORF2 proteins bind L1 RNAs, forming ribonucleoproteins (RNPs). These RNPs interact with diverse host proteins, some repressive and others required for the L1 lifecycle. Using differential affinity purifications and quantitative mass spectrometry, we have characterized the proteins associated with distinctive L1 macromolecular complexes. Our findings support the presence of multiple L1-derived retrotransposition intermediates in vivo. Among them, we describe a cytoplasmic intermediate that we hypothesize to be the canonical ORF1p/ORF2p/L1-RNA-containing RNP, and we describe a nuclear population containing ORF2p, but lacking ORF1p, which likely contains host factors participating in template-primed reverse transcription.


Introduction
Sequences resulting from retrotransposition constitute more than half of the human genome and are considered to be major change agents in eukaryotic genome evolution (Kazazian, 2004). L1 retrotransposons have been particularly active in mammals (Furano et al, 2004), comprising ~20% of the human genome (Lander et al, 2001); somatic retrotransposition has been widely implicated in cancer progression (Lee et al, 2012;Tubio et al, 2014) and may even play a role in neural development (Muotri et al, 2005). Despite the magnitude of their contributions to mammalian genomes, L1 genes are modest in size.
A full-length L1 transcript is ~6 knt long and functions as a bicistronic mRNA that encodes two polypeptides, ORF1p and ORF2p (Ostertag & Kazazian, 2001), which respectively comprise a homotrimeric RNA binding protein with nucleic acid chaperone activity (Martin & Bushman, 2001) and a multifunctional protein with endonuclease and reverse transcriptase activities (Mathias et al, 1991;Feng et al, 1996). Recently, a primate-specific third ORF, named ORF0, has been identified on the Crick strand of the L1 gene; this ORF encodes a 71 amino acid peptide and may generate insertion site-dependent ORFs via splicing (Denli et al, 2015). ORF1p and ORF2p are thought to interact preferentially with the L1 RNA from which they were translated (in cis), forming a ribonucleoprotein (RNP) (Kulpa & Moran, 2006;Taylor et al, 2013) considered to be the canonical direct intermediate of retrotransposition (Hohjoh & Singer, 1996;Kulpa & Moran, 2005;Martin, 1991;Kulpa & Moran, 2006;Doucet et al, 2010). L1s also require host factors to complete their lifecycle (Suzuki et al, 2009;Peddigari et al, 2013;Dai et al, 2012;Taylor et al, 2013) and, consistent with a fundamentally parasitic relationship (Beauregard et al, 2008), the host has responded by evolving mechanisms that suppress retrotransposition (Goodier et al, 2013;Arjan-Odedra et al, 2012;Goodier et al, 2012;Niewiadomska et al, 2007). It follows that as the host and the parasite compete, L1 expression is likely to produce a multiplicity of RNP forms engaged in discrete stages of retrotransposition, suppression, or degradation.
Although L1 DNA sequences are modestly sized compared to typical human genes, L1 intermediates are nevertheless RNPs with a substantially sized RNA component; e.g. larger than the ~5 knt 28S rRNA (Gonzalez et al, 1985) and approximately three to four times the size of a "typical" mRNA transcript (Lander et al, 2001;Sommer & Cohen, 1980). Therefore, it is likely that many proteins within L1 RNPs form interactions influenced directly and indirectly by physical contacts with the L1 RNA. We previously determined that L1 RNA comprised ~20-30% of mappable RNA-seq reads in ORF2p-3xFLAG affinity captured fractions (Taylor et al, 2013). We also observed that the retention of ORF1p and UPF1 within affinity captured L1 RNPs was reduced by treatment with RNases (Taylor et al, 2013). In the same study we observed that two populations of ORF2p-associated proteins could be separated by split-tandem affinity capture (ORF2p followed by ORF1p), a twodimensional affinity enrichment procedure (Caspary et al, 1999;Taylor et al, 2013). Initial characterization of these two L1 populations by western blotting suggested that discrete L1 populations were likely primed for function in different stages of the lifecycle. We therefore expected additional uncharacterized complexity in the spectrum of L1-associated complexes present in our affinity purified fractions.
In this study, we have used quantitative mass spectrometry (MS) to investigate the proteomic characteristics of endogenously assembled ectopic L1 complexes present in an assortment of affinity-enriched fractions. We revisited RNase treatment and split-tandem affinity capture approaches and complemented them with in-cell localization of ORF proteins by immunofluorescence microscopy (see also the companion manuscript by Mita et al. [co-submission]). We additionally explored L1 proteomes associated with catalytically-inactivated ORF2p point mutants and monitored the rates of protein exchange from L1s in vitro. Taken together, our data support the existence of a variety of putative L1-related protein complexes.

Results
Affinity proteomic experiments conducted in this study use quantitative MS based upon metabolic labeling (Oda et al, 1999). Two main experimental designs (and modifications thereof) facilitating quantitative cross-sample comparisons have been used: SILAC (Ong et al, 2002;Wang & Huang, 2008) and I-DIRT (Tackett et al, 2005;Taylor et al, 2013). In these approaches, cells are grown for several doublings in media containing amino acids composed either of naturally-occurring 'light' isotopes or biologically identical 'heavy' isotopes (e.g. 13 C, 15 N lysine and arginine), such that the proteomes are thoroughly labeled.
Protein fractions derived from the differently labeled cell populations, obtained e.g. before and after experimental manipulations are applied, are mixed and the relative differences in proteins contributed by each fraction are precisely measured by mass spectrometry. In addition to the above cited studies, these approaches have been adapted to numerous biological questions using a variety of analytical frameworks e.g. (Byrum et al, 2011;Luo et al, 2016;Trinkle-Mulcahy et al, 2008;Ohta et al, 2010;Kaake et al, 2010;Geiger et al, 2011). Because it is challenging to speculate on the potential physiological roles of protein interactions that form after extraction from the cell, we often use I-DIRT, which allows the discrimination of protein-protein interactions formed in-cell from those occurring postextraction. Our prior affinity proteomic study, based on I-DIRT, identified 37 putative in vivo interactors (Taylor et al, 2013), described in Table 1. Therefore, primarily focus here on the behaviors of these "I-DIRT significant" L1 interactors, in order to determine their molecular associations and ascertain the variety of distinctive macromolecular complexes formed in-cell that co-purify with affinity-tagged ORF2p. The complete lists of proteins detected in each experiment are presented in the supplementary information (see Table   S1). Except where noted otherwise, the presented experiments were conducted in suspension-cultured HEK-293T LD cells, using a synthetic L1 construct -ORFeus-HS -driving the expression 3xFLAG-tagged L1 (ORF1; ORF2::3xFLAG; 3'-UTR) from a tetracycline inducible minimal-CMV promoter, harbored on a mammalian episome (pLD401 (Taylor et al, 2013;An et al, 2011;Dai et al, 2012)). All purified L1s described in this study were obtained by affinity capture of ORF2p-3xFLAG before further experimental manipulations were applied. We have represented any ambiguous protein group, which occurs when the same peptides identify a group of homologous protein sequences, with a single, consistently applied gene symbol and a superscript 'a' in all figures. Table S1 contains the references to other proteins explaining the presence of the same peptides. For example, RPS27A, (ubiquitin) UBB, UBC, and (ribosomal Protein L40) UBA52 can be explained by common ubiquitin peptides shared by these genes. RPS27A-specific peptides were not identified in this study, but we retained the nomenclature for consistency with our previous work; HSPA1A is reported in this study, but cannot be distinguished from the essentially identical protein product of HSPA1B.

RNase-sensitivity exhibited by components of affinity purified L1 RNPs
Figure 1 (panels A-C) illustrates the approach and displays the findings of our assay designed to reveal which proteins depend upon the presence of intact L1 RNA for retention within the purified L1 RNPs. Briefly, metabolically-labeled affinity captured L1s were treated either with a mixture of RNases A and T1 -thus releasing proteins that require intact RNA to remain linked to ORF2p and the affinity medium -or BSA, as an inert control. After removing the fractions released by the RNase or BSA treatments, the proteins remaining on the affinity media were eluted with lithium dodecyl sulfate (LDS), mixed together, and then analyzed by MS. Proteins released, and so depleted, by RNase treatment were thus found to be more abundant in the BSA-treated control. The results obtained corroborate and extend our previous findings: ORF1p and UPF1 exhibited RNase-sensitivity (Taylor et al, 2013). We also observed that ZCCHC3 and MOV10 exhibited RNasesensitivity to a level similar to ORF1p. The remaining I-DIRT significant proteins were RNase-resistant in this assay. With the exception of the PABPC1/4 proteins (and ORF2p itself, see Discussion), the I-DIRT significant proteins (colored nodes, Fig. 1C) that were resistant to RNase treatment (nearest the origin of the graph) classify ontologically as nuclear proteins (GO:0005634, p ≈ 3 x 10 -4 , see Methods). These same proteins were previously observed as specific L1 interactors in I-DIRT experiments targeting ORF2p but not in those targeting ORF1p; in contrast, the proteins that demonstrated RNase-sensitivity: ORF1, MOV10, ZCCHC3, and UPF1 were observed in both ORF1p and ORF2p I-DIRT experiments ( Table 1). Stated another way, the proteins released upon treating an affinity captured ORF2p fraction with RNases are among those that can also be obtained when affinity capturing ORF1p directly, while those that are RNase-resistant are not ORF1p interactors (Taylor et al, 2013). The ORF1p-linked, I-DIRT significant, RNase-sensitive proteins were too few to obtain a high confidence assessment of ontological enrichment; but, when combined with remaining proteins exhibiting sensitivity to the RNase treatment (black nodes, Fig. 1C), they together classified as 'RNA binding' (GO:0003723, p ≈ 1 x 10 -11 ). This analysis also revealed a statistically significant overrepresentation of genes associated with the exon junction complex (EJC, GO: 0035145, p ≈ 1 x 10 -6 , discussed below). ORF1p/2p-associated interactomes appear to overlap on the L1 RNA, as expected.
However, considering the nuclear ontology associated with the RNase-resistant ORF2p interactors, we conclude that ORF1p probably does not interact with ORF2p in the nucleus, where the ORF2p interactome is either not broadly dependent upon L1 RNA, or is protected from RNases. ORF1p may be displaced from the L1 RNA during the nuclear portion of the lifecycle. Thus, host-encoded proteins appear to segregate themselves into groups according to their L1 ORF protein associations, and that these groups respond differently, yet in an apparently coordinated fashion, to RNase treatment. This observation led to the hypothesis that our ORF2p-3xFLAG affinity purified L1s constitute a composite purification of at least, but not limited to, (1) a cytoplasmic population of L1-RNA-dependent, ORF1p/ORF2pcontaining L1 RNPs, and (2) an ORF1p-independent nuclear population associated with ORF2p.
While effects of PABPC1, MOV10, and UPF1 on L1 activity have been described (Arjan-Odedra et al, 2012;Taylor et al, 2013;Dai et al, 2012), effects of ZCCHC3 on L1 remained uncharacterized. ZCCHC3 is an RNA-binding protein associated with poly(A)+ RNAs (Castello et al, 2012) but otherwise little is known concerning its functions. Notably, in a genome-wide screen, small interfering (si)RNA knockdown of ZCCHC3 was observed to increase the infectivity of the Hepatitis C, a positive sense RNA virus (Li et al, 2009); and ZCCHC3 was observed to copurify with affinity captured HIV, a retrovirus, at a very high SILAC ratio (>10), supporting the specificity of this interaction (Engeland et al, 2014). We therefore explored the effects on L1 mobility both of over-expression and siRNA knockdown of ZCCHC3. Over-expression of ZCCHC3 reduced L1 retrotransposition to ~10% that observed in the control, consistent with a negative regulatory role for ZCCHC3 in the L1 lifecycle; small interfering RNA (siRNA) knockdown of ZCCHC3 induced a modest increase in retrotransposition compared to a scrambled control siRNA (~1.9x ±0.1; Table S2).
Moreover, although not among our I-DIRT hits (see Discussion), the presence of EJC components (MAGOH, RBM8A, EIF4A3, UPF1) among the RNase-sensitive fraction of proteins intrigued us, given that L1 genes are intronless. We speculated that L1s may use EJCs to enhance nuclear export, evade degradation by host defenses, and/or aggregate with mRNPs within cytoplasmic granules. For this reason we carried out a series of siRNA knockdowns of these EJC components and other physically or functionally related proteins found in the affinity purified fraction (listed in Table S2). siRNA knockdowns of RBM8A and EIF4A3 caused inviability of the cell line. We found that knocking-down MAGOH and the EJC-linked protein IGF2BP1 (Jønson et al, 2007) both, respectively, reduced retrotransposition by ~50%, consistent with a role in L1 proliferation; although, these knockdowns also caused a reduction in viability of the cell line (See Discussion).

Split-tandem separation of compartment-specific L1 ORF-associated complexes
To further test our hypothesis and better characterize the components of our L1 fraction, we conducted split-tandem affinity capture. Figure 1 (panels E-F) illustrates the approach and displays the findings of the assay, which physically separated ORF1p/ORF2pcontaining L1 RNPs from a presumptive 'only-ORF2p-associated' population. Briefly, metabolically-labeled L1s were affinity captured by ORF2p-3xFLAG (1 st dimension) and the obtained composite was subsequently further fractionated by α-ORF1p affinity capture (2 nd dimension, or split-tandem purification), resulting in α-ORF1p-bound and unbound fractions. The bound fraction was eluted from the affinity medium with LDS. The bound and unbound fractions were then mixed and analyzed by MS to ascertain proteomic differences between them. The fraction eluted from the α-ORF1p medium contained the population of proteins physically linked to both ORF2p and ORF1p, whereas the supernatant from the α-ORF1p affinity capture contained the proteins associated only with ORF2p (and, formally, those which have dissociated from the ORF1p/ORF2p RNP). The results corroborated our previous observations that: i) almost all of the ORF1p partitioned into the elution fractions, ii) a quarter of the ORF2p (~26%) followed ORF1p during the α-ORF1p affinity capture, iii) roughly half of the UPF1 (~55%) followed ORF1p, and iv) most of the PCNA (~87%) remained in the ORF1p-depleted supernatant fraction (Fig. 1F, and consistent with prior estimates based on protein staining and western blotting (Taylor et al, 2013)); thus v) supporting the existence of at least two distinct populations of L1-ORFprotein-containing complexes in our affinity purifications.
The population eluted from the α-ORF1p affinity medium (Fig. 1D, far right gel lane,and nodes located in the upper right of the graph, panel F) is consistent with the composition of the ORF1p/ORF2p-containing L1 RNP suggested above. Our split-tandem separation segregated the constituents of the purified L1 fraction comparably to the RNase-sensitivity assay, both in terms of which proteins co-segegate with ORF1p/ORF2p (compare Fig. 1C and F, blue nodes, upper right of graphs) as well as those which appear to be linked only to  Table 1). Analysis of the nodes whose degree of ORF1p association was similar to that of UPF1 (blue nodes exhibiting ≥ 55% ORF1p co-partitioning, Fig. 1F) revealed that they map ontologically to a 'cytoplasmic ribonucleoprotein granule' classification (GO:0036464, p ≈ 6 x 10 -8 ; see Discussion). In contrast, all sixteen proteins exhibiting ORF1p co-partitioning approximately equal to or less than that of ORF2p were predominantly found in the supernatant fraction and were enriched for cell-compartment-specific association with the nucleus (GO:0005634, p ≈ 4 x 10 -5 ; Fig. 1F, all magenta nodes below PARP1, inclusive, ≤ 36%). These two fractions therefore appear to be associated with different cell compartments, reaffirming our postulate: the ORF1p/ORF2p-containing population is a cytoplasmic intermediate related to the canonical L1 RNP typically ascribed to L1 assembly in the literature, and the predominantly ORF2p-associated population comprises a putative nuclear interactome; each therefore referred to, respectively, as cytoplasmic and nuclear L1 interactomes hereafter.
From the same analysis, we noted that PURA, PURB, PCNA, and TOP1 which all partition predominantly with nuclear L1, exhibited an ontological co-enrichment (termed 'nuclear replication fork,' GO:0043596, p ≈ 3 x 10 -4 ). The nodes representative of PURA, PURB, and PCNA appeared to exhibit a striking proximity to one another, suggesting highly similar cofractionation behavior potentially indicative of direct physical interactions. In an effort to examine this possibility, we graphed the frequency distribution of the proximities of all three-node-clusters observed within Figure 1F, revealing the likelihood of the PURA/PURB/PCNA cluster to be p = 3.2 x 10 -7 (see Supplementary Methods). We therefore concluded that PURA, PURB, PCNA, and (perhaps at a lower affinity) TOP1, likely constitute a physically associated functional module interacting with L1. In further support of this assertion, we noted that known functionally linked protein pairs PABPC1/PABPC4 (cytoplasmic) (Jønson et al, 2007;Katzenellenbogen et al, 2007) and HSPA8/HSPA1A (nuclear) (Jønson et al, 2007;Nellist et al, 2005) also exhibited comparable co-partitioning by visual inspection, and statistical testing of these clusters revealed the similarity of their co-partitioning to be significant at p ≈ 0.001 for the former, and p ≈ 0.0002 for the latter.
The observed variation in co-partitioning behavior between the different proteins comprising the nuclear L1 fraction might reflect the presence of multiple distinctive (sub)complexes present within this population.
To validate our hypothesis that these proteins are associated with ORF2p in the nucleus, we carried out ORF2p-3xFLAG affinity capture from chromatin-enriched sub-cellular fractions and found that the co-purified proteins we identified largely overlapped with those described above (Table S3), including: PARP1, PCNA, UPF1, PURA, and TOP1. Although we observed that silencing PCNA expression adversely affects L1 retrotransposition (Taylor et al, 2013), we found that knocking down TOP1 approximately doubled retrotransposition frequency, while a more modest 1.4x increase effect was observed for PURA, and no substantial effect was observed for PURB, compared to a scrambled siRNA control. In contrast, over-expression of PURA reduced retrotransposition to ~20% of the expected level (Table S2).

ORF1p/ORF2p immunofluorescence protein localization
Although our proteomic and biochemical analyses supported the existence of distinctive nuclear and cytoplasmic L1 populations, our prior immunofluorescence (IF) analyses did not reveal an apparent nuclear population, leading us to revisit IF studies. Previously, IF of ORF1p and ORF2p in HeLa and HEK-293T cells yielded two striking observations: i) ORF2 expression was seemingly stochastic, with ORF2p observed in ~30% of cells; and ii) while ORF1p and ORF2p co-localized in cells that exhibited both, we did not observe an apparent nuclear population of either protein (Taylor et al, 2013). Subsequently, we noted an absence of mitotic cells from these preparations. Reasoning that these cells were lost due to selective adherence on glass slides, and noting that cell division has been reported to be promote L1 transposition (Xie et al, 2013;Shi et al, 2007), we repeated the assays using puromycin-selected Tet-on HeLa cells grown on fibronectin coated coverslips. The results are shown in Figure 2.
The modified IF assay corroborated our prior results in that nearly all the cells exhibited cytoplasmic ORF1p and ~30% also exhibited cytoplasmic ORF2p, which co-localized with ORF1p ( Fig. 2, top row). However, a new population of cells also became apparent, consisting of pairs of cells exhibiting nuclear localized ORF2p (Fig. 2, middle row). These cells accounted for <10% of the population and, because of they always occurred in proximal pairs, are presumed to have recently gone through mitosis. Expression of ORF2 in the absence of ORF1 (ΔORF1; pLD561) resulted in >90% of cells exhibiting cytoplasmic ORF2p, consistent with our previous work (Taylor et al, 2013). We did not observe instances of nuclear ORF2p using the construct (Fig. 2

, bottom row), suggesting that
ORF1p is required for ORF2p nuclear localization. In a separate study, including more detailed analyses ORF protein localization, Mita et al. (co-submission) observed that both ORF proteins enter the nucleus of HeLa cells during mitosis, however, nuclear ORF1p does not seem to be physically associated with nuclear ORF2p (see Discussion). Taken together, the data obtained from the modified IF experiments aligned well with our proteomic and biochemical data: L1 expression resulted in at least two distinct populations: cytoplasmic complexes containing both ORF1p and ORF2p, and nuclear complexes containing ORF2p while potentially lacking ORF1p.

The effects of retrotransposition-blocking point mutations on the interactomes of purified L1 RNPs
Based on the hypothesis that our composite purifications contain bona fide nuclear intermediates, we decided to explore the effects of catalytic point mutations within the ORF2p endonuclease and reverse transcriptase domains, respectively. We reasoned that such mutants may bottleneck L1 intermediates at the catalytic steps associated with host gDNA cleavage and L1 cDNA synthesis, potentially revealing protein associations that are important for these discrete aspects of template primed reverse transcription (TPRT), the presumed mechanism of L1 transposition (Luan et al, 1993;Feng et al, 1996;Cost et al, 2002). For this we used an H230A mutation to inactivate the endonuclease activity (EN -/ pLD567), and a D702Y mutation to inactivate the reverse transcriptase activity (RT -/ pLD624) (Taylor et al, 2013). Figure 3 illustrates the approach and displays the findings of our assay. Broadly, our findings revealed several classes of behaviors (Fig. 3B). Two classes of behavior appeared to be particularly striking: (1) the yield of constituents of cytoplasmic L1s was reduced, relative to WT, by the EN mutation, yet elevated by the RT mutation ( Fig.   3B, left side); and (2) numerous constituents of nuclear L1s were elevated in yield by the EN mutation but reduced or nominally unchanged, relative to WT, by the RT mutation (Fig.

3B, right side). While reductions in the levels of cytoplasmic L1 proteins within EN -L1
RNPs may be difficult to rationalize, accumulation of these proteins within RT -L1s could be explained by a decrease in the frequency that ORF2p traverses the L1 RNA, normally displacing these proteins to some degree even within the cytoplasm. With respect to the second group, IPO7, NAP1L4, NAP1L1, FKBP4, HSP90AA1, and HSP90AB1 were all elevated in the ENmutants, potentially implicating these proteins as part of an L1 complex (or complexes) immediately preceding DNA cleavage. Notably, there is a third class of proteins, including PURA/B, PCNA, TOP1, and PARP1, that all respond similarly to both ENand RTmutants compared to WT, exhibiting reduced associations with the mutant L1s; although, the RTmutant showed a larger effect size on the PURA/B proteins. These data suggest that cleavage of the host genomic DNA fosters associations between L1 and this group of proteins, but that interactions with PURA/B are further enhanced by L1 cDNA production. Other nuclear L1 proteins: HSPA8, HAX1, HSPA1A, TUBB, and TUBB4B were increased in both mutants. To better visualize the range of behaviors exhibited by our proteins of interest, and the population at large, we cross-referenced the relative enrichments of each protein detected in both experiments, shown in Figure 3C. We noted that the same striking trend mentioned above, two seemingly opposite behavioral classes of proteins could also be observed, globally among all proteins (see Fig 3B, left side and right side, and Fig. 3C). The crisscross pattern displayed in (and Fig. S2) indicates that two largely distinctive proteomes were purified with these two classes of mutants. Notably, the pattern observed appears to track with the relative behavior of ORF1p, which, along with other cytoplasmic L1 factors is elevated in RTmutants and reduced in ENmutants. We therefore speculate that the sum of observed interactomic changes include effects attributable directly to the catalytic mutations as well as indirect effects resulting from the response of ORF1p to the mutations.

Dynamics of L1 RNPs in vitro
We next decided to measure the in vitro dynamics of proteins co-purifying with affinity captured L1s, reasoning that proteins with comparable profiles are likely candidates to be physically linked to one another or otherwise co-dependent for maintaining stable interactions with L1s. To achieve this, we first purified heavy-labeled, affinity-tagged L1s and subsequently incubated them, while immobilized on the medium, with light-labeled, otherwise identically prepared cell extracts from cells expressing untagged L1s (Luo et al, 2016). In this scenario, heavy-labeled proteins present at the zero time point are effectively "infinitely diluted" with light-labeled cell extract. The exchange of proteins, characterized by heavy-labeled proteins decaying from the immobilized L1s and being replaced by lightlabeled proteins supplied by the cell extract, was monitored by quantitative MS. These experiments were conducted using constructs based on the naturally occurring L1 RP sequence (Dai et al, 2014;Taylor et al, 2013;Kimberland et al, 1999). Figure 4 illustrates the approach and displays the findings of our assay. We observed three distinctive clusters of behaviors (Fig. 4B, C). Notably, ORF1p, ZCCHC3, and the cytoplasmic poly(A) binding proteins clustered together, forming a relatively stable core complex. Exhibiting an intermediate level of relative in vitro dynamics, UPF1 and MOV10 clustered with TUBB, TUBB4B, and HSP90AA1. A third, and least stable, cluster consisted of only nuclear L1 interactors.

Multidataset Integration
Having observed coordinated and distinctive behaviors exhibited by groups of L1 interacting proteins across several distinctive biochemical assays, we then integrated the data and calculated the behavioral similarity of the I-DIRT-significant interactors, producing a dendrogram; Figure 5 displays their relative similarities. A cluster containing the putative cytoplasmic L1 components (MOV10, UPF1, ZCCHC3, PABC1/4, ORF1p) was observed, as was a cluster containing PURA/B, PCNA, TOP1, PARP1, aligning with our assessments of the separated datasets (Figs. 1, 3, 4). In addition to these, we also observed three distinctive clusters derived from the nuclear L1 interactome. We believe that this is likely to reflect the presence of a collection of distinctive macromolecules.

Discussion
In this study we have characterized biochemical, interatomic, enzymatic, and cellular localization properties of ectopically expressed L1s. Through the assays explored, we observed discrete and coordinated behaviors, permitting us to refine our model of L1  -submission), which also present data to support the hypothesis that PCNA-associated ORF2p is not appreciably associated with ORF1p, and also identified TOP1 and PARP1 in complex with ORF2p/PCNA.

Cytoplasmic L1s
ORF1p, MOV10, UPF1, and ZCCHC3 are released from L1 RNPs by treatment with RNases ( Fig. 1), indicating the importance of the L1 RNA in the maintenance of these interactions.
In this context, the L1 ORF and poly(A) binding proteins support L1 proliferation (Kulpa & Moran, 2006;Dai et al, 2012;Wei et al, 2001), whereas ZCCHC3 (Table S2) and MOV10 (Goodier et al, 2012;Arjan-Odedra et al, 2012) function in repressive capacities. Although UPF1 might also be expected to operate in a repressive capacity through its role in nonsense mediated decay (NMD), we previously demonstrated that UPF1's role does not apparently resemble that of canonical NMD and it acts as an enhancer of retrotransposition despite negatively affecting L1 RNA and protein levels, supporting the possibility of repressive activity in the cytoplasm and proliferative activity in the nucleus (Taylor et al, 2013). Notably, MOV10 has been implicated in the recruitment of UPF1 to mRNA targets through protein-protein interactions (Gregersen et al, 2014). However, we observed that MOV10 exhibited a greater degree of RNase-sensitivity than UPF1, indicating that, if MOV10 directly modulates the UPF1 interactions with L1, a sub-fraction of UPF1 exhibits a distinct behavior (UPF1 is ~62% as sensitive to RNase treatment as MOV10, Fig. 1C).
Bimodal UPF1 behavior can also be seen in split-tandem capture experiments, and UPF1 was recovered with L1s affinity purified from fractionated chromatin (further discussed below), and only about half of the UPF1 exhibits ORF1p-like partitioning with the canonical L1 RNP (Fig. 1F). Presumably, the RNase-sensitive fraction, released in concert with MOV10, is the same fraction observed in cytoplasmic L1s obtained by split-tandem purification. In contrast, PABPC1 and C4 exhibit strong ORF1p-like partitioning (comparable to MOV10), but appear wholly insensitive to RNase treatment. This is most likely due to the fact that neither RNase A nor T1 cleave RNA at adenosine residues (Volkin & Cohn, 1953;Yoshida, 2001); hence poly(A) binding proteins may not be ready targets for release from direct RNA binding by the assay implemented here (or generally, using these ribonucleases). Failure to release of ORF2p into the supernatant upon RNase treatment is expected due to its immobilization upon the affinity medium (Dai et al, 2014). However, we note that ORF2p binding to the L1 RNA has also been proposed to occur at the poly(A) tail (Doucet et al, 2015), raising the related possibility of close physical association on the L1 RNA between ORF2p and PABPC1/4 in cytoplasmic L1 RNPs. ORF1p, PABPC1/4, MOV10, ZCCHC3, and UPF1, all behaved comparably in response to ENand RTcatalytic mutations, decreasing together in ENmutants, and increasing together in RTmutants (Fig.   3B). Moreover, when the exchange of proteins within L1 RNPs was monitored directly, PABPC1/4 and ZCCHC3 exhibited nearly identical stability, well above the background distribution; UPF1 and MOV10 also exhibited comparable kinetics to one another, falling into an intermediate stability cluster (Fig. 4B, C).
RNase-sensitivity was displayed by numerous proteins not previously identified as putative L1 interactors (Table 1, Fig. 1 and (Taylor et al, 2013)). A known limitation of I-DIRT (and many SILAC-based analyses) is that it cannot discriminate non-specific interactors from specific but rapidly exchanging interactors (Wang & Huang, 2008;Luo et al, 2016;Smart et al, 2009). Our samples likely contain rapidly exchanging, physiologically relevant factors that were not revealed by I-DIRT under the experimental conditions used. With this in mind, we note members of the exon junction complex (EJC), RBM8A (Y14), EIF4A3 (DDX48), and MAGOH, are among our RNase-sensitive constituents, with all exhibiting a similar degree of RNase-sensitivity (Fig. 1C, labeled black dots). Crucially, these proteins are physically and functionally connected to UPF1 (reviewed in (Schweingruber et al, 2013)), and physically to MOV10 (Gregersen et al, 2014), both validated L1 interactors.
We therefore hypothesize that EJCs may constitute bona fide L1 interactors missed in our original screen. This may seem unexpected because canonical L1 RNAs are thought not to be spliced, but this assumption has been challenged by one group (Belancio et al, 2006), and splicing-independent recruitment of EJCs has also been demonstrated (Budiman et al, 2009). Perhaps more compelling, EJC proteins exhibited a striking similarity in RNasesensitivity to MOV10 (Fig. 2B). EIF4A3 has been suggested to form an RNA-independent interaction with MOV10 (Gregersen et al, 2014), and MOV10 is a known negative regulator of L1, making it attractive to speculate that these proteins were recruited and released in concert with MOV10 and/or UPF1.
Ectopically expressed canonical L1 RNPs have been shown to accumulate in cytoplasmic stress granules (Doucet et al, 2010;Goodier et al, 2010), and our observation of UPF1, MOV10, and MAGOH in the RNase-sensitive fraction is consistent with this characterization (Jain et al, 2016). However, the additional presence of EIF4A3 and RBM8A suggested that our RNPs may instead overlap with IGF2BP1 (IMP1) granules, reported to be distinct from stress granules (Jønson et al, 2007;Weidensdorfer et al, 2009). Consistent with this possibility, we observed IGF2BP1, YBX1, DHX9, and HNRNPU within the mixture of copurified proteins (Table S1). We did not, however, observe canonical stress granule markers G3BP1 or TIA1 (Goodier et al, 2007;Jain et al, 2016;Doucet et al, 2010).
Surprisingly, siRNA knockdown of IGF2BP1 substantially reduced L1 retrotransposition; however, we note that the cytotoxicity associated with knocking-down EJC components may confound interpretation (Table S2). Given the result obtained, IGF2BP1 appears to support L1 proliferation. Consistent with an established function (Bley et al, 2015;Weidensdorfer et al, 2009), IGF2BP1 granules may sequester and stabilize L1 RNPs in the cytoplasm, creating a balance of L1 supply and demand that favors proliferation over degradation. Although human L1 does not contain a known IRES, it is known that ORF2 is translated by a non-canonical mechanism (Alisch et al, 2006), and IGF2BP1 may promote this (Weinlich et al, 2009).

Nuclear L1s
The fraction eluted from the α-ORF1p medium contained the population of proteins physically linked to both ORF2p and ORF1p and greatly resembled the components released upon RNase treatment, hence these linkages primarily occur through the L1 RNA (or are greatly influenced by it). In contrast, the supernatant from the α-ORF1p affinity capture contained the proteins we speculate to be associated with ORF2p, but not ORF1p; moreover, fully intact RNA does not appear to be essential to the maintenance of these interactions. Notable within the latter group of interactors was the PURA/PURB/PCNA cluster, with TOP1 also in close proximity, ontologically grouping to the nuclear replication fork (GO:0043596). Separately, a few physical and functional connections have been shown for PURA/PURB (Knapp et al, 2006;Kelm et al, 1999;Mittler et al, 2009), PCNA/TOP1 (Takasaki et al, 2001), and PURA/PCNA (Qin et al, 2013). Notably, PURA, PURB, and PCNA have been independently linked to replication-factor-C / replicationfactor-C-like clamp loaders (Kubota et al, 2013;Havugimana et al, 2012). Given that we also observe tight clustering of protein pairs known to be physically and functionally linked, e.g. PABPC1/4 (Jønson et al, 2007;Katzenellenbogen et al, 2007) and HSPA8/1A (Jønson et al, 2007;Nellist et al, 2005), and because we have established PCNA as a positive regulator of L1 retrotransposition (Taylor et al, 2013), we propose that the [PURA/B/PCNA/TOP1] group is a functional sub-complex of nuclear L1. In addition, although it does not cluster as closely to the [PURA/B/PCNA/TOP1] group, PARP1 is found within the putative nuclear L1 population and is functionally linked with PCNA, specifically stalled replication forks (Bryant et al, 2009;Min et al, 2013;Ying et al, 2016). Further tying them together, these proteins all also exhibited affinity capture yield decreases of severalfold in response to mutations that abrogated ORF2p EN or RT activity (Fig. 3). This is compelling because these ORF2p enzymatic activities are required in order for it to to manipulate DNA and traverse the steps of the L1 lifecycle that benefit from physical association with replication forks. One caveat to this interpretation is that, while knocking down PCNA reduced L1 retrotransposition (Taylor et al, 2013), no such effect was observed for TOP1 or PURA/B, which led instead to mild increases in L1 activity (Table S2). These proteins may be physically assembled within a common intermediate, but functionally antagonistic. HSP90 proteins were also observed in this fraction, and are also linked with stalled replication forks (Arlander et al, 2003;Ha et al, 2011), but exhibited a distinctive response to catalytic mutants, accumulating in ENmutants while exhibiting a modest decrease in RTmutants. The recruitment of the ORF2p/PCNA complex to stalled replication forks has been also independently proposed by Mita et al. (co-submission).
As mentioned above, we previously speculated that an RNase-insensitive fraction of L1associated UPF1 may support retrotransposition in conjunction with PCNA in the nucleus ( (Azzalin & Lingner, 2006;Taylor et al, 2013)

and Mita et al. [co-submission]). In contrast
to other PCNA-linked proteins, catalytic inactivation of ORF2p did not robustly affect the relative levels of co-purified UPF1, and UPF1 behaved in a distinct manner during tandem purification. The equivocal behavior of UPF1 in several assays (Figs. 1, 3, and 4) supports UPF1's association with both the putative cytoplasmic and nuclear L1 populations, the latter being additionally supported by the association of UPF1 with ORF2p-3xFLAG purified from chromatin (Table S3). NAP1L4, NAP1L1, FKBP4, HSP90AA1, and HSP90AB1 (Baltz et al, 2012;Castello et al, 2012;Simon et al, 1994;Rodriguez et al, 1997;Peattie et al, 1992) are associated with RNA binding, involved in protein folding and unfolding, and function as nucleosome chaperones. An interesting possibility is that they have a nucleosome remodeling activity that may be required to allow reverse transcription to begin elongating efficiently, or for assembly of nucleosomes on newly synthesized DNA.

Future Studies
An obvious need is the continued validation of putative interactors by in vivo assays. RNAi coupled with L1 insertion measurements by GFP fluorescence (Ostertag et al, 2000) provides a powerful method to detect effects on L1 exerted by host factors. However, this approach can sometimes be limited by cell viability problems associated with important genes; it is therefore critical to control for this (Table S2). Along these lines, a high throughput approach has been developed for genome-wide screening of L1-host protein interactions (P. Mita and J. Boeke, manuscript in preparation). Throughout this and our prior study (Taylor et al, 2013) we have used comparable in vitro conditions for the capture and analysis of L1 interactomes. However, we are aware that this practice has enforced a single biochemical "keyhole" through which we have viewed L1-host protein associations. It is important to expand the condition space in which we practice L1 interactome capture and analysis in order to expand our vantage point on the breadth of L1-related macromolecules (Hakhverdyan et al, 2015). In concert with this, we must develop sophisticated, automated, reliable, low-noise methods to integrate biochemical, proteomic, genomic, and ontological data; the first stages of which we have attempted in the present study. Although we have used I-DIRT to increase our chances of identifying bona fide interactors (Tackett et al, 2005;Taylor et al, 2013), it is clear, and generally understood, that some proteins not making the significance cut-off will nevertheless prove to be critical to L1 activity (Byrum et al, 2011;Luo et al, 2016;Joshi et al, 2013), such as demonstrated by our unexpected findings with IGF2BP1 (Table S2). Through further development, including reliable integration with diverse, publicly available interactome studies, we hope to enable the detection of extremely subtle physical and functional distinctions between (sub)complexes and their components, considerably enhancing reliable exploration and hypothesis formation. Furthermore, it is striking that no structures of assembled L1s yet exist; these are missing data that are likely to provide a profound advance for the mechanistic understanding of L1 molecular physiology. However, we believe that with the methods presented here, endogenously assembled ORF1p/ORF2p/L1-RNA-containing cytoplasmic L1 RNPs can be prepared at sufficiently high purity and yield Fig. 1F) to enable electron microscopy studies. Importantly, we have shown that our affinity purified fractions are enzymatically active for reverse transcription of the L1 RNA (Taylor et al, 2013) and endonucleolytic cleavage of DNA (J. LaCava, unpublished observation), providing some hope that cryo-electron microscopy could be used to survey the dynamic structural conformations of L1s formed during its various lifecycle stages (Takizawa et al, 2017).

Methods
The preparation of L1 RNPs was carried out essentially as previously described (Taylor et al, 2013(Taylor et al, , 2016, with modifications described here. Briefly, HEK-293T LD cells (Dai et al, 2012) transfected with L1 expression vectors were cultured as previously described or using a modified suspension-growth SILAC strategy described below. L1 expression was induced with with 1μg / ml doxycycline for 24 hours, and the cells were harvested and extruded into liquid nitrogen. In all cases the cells were then cryogenically milled  and used in affinity capture experiments and downstream assays.

Modified SILAC Strategy
Freestyle-293 medium lacking Arginine and Lysine was custom-ordered from Life Technologies, and heavy or light amino acids plus proline were added at the same concentrations previously described (Taylor et al, 2013), without antibiotics. Suspensionadapted HEK-293T LD were spun down, transferred to SILAC medium and grown for >7 cell divisions in heavy or light medium. On day 0, four (4) 1L square glass bottles each containing 200 ml of SILAC suspension culture at ~2.5×10 6 cells/ml were transfected using 1 μg/ml DNA and 3 μg/ml polyethyleneimine "Max" 40 kDa (Polysciences, Warrington, PA, #24765). A common transfection mixture was made by pre-mixing 800 μg DNA and 2.4mL of 1 mg/ml PEI-Max in 40 ml Hybridoma SFM medium (Life Technologies, Grand Island, NY, #12045-076) and incubating for 20 min at room temperature (RT); 10 ml of the mixture was added to each bottle. On day 1, cells (200 ml) were split 1:2.5 (final two bottles each containing 250 mL) without changing the medium. On day 3, the cells were induced with 1 µg/ml doxycycline, and on day 4 the cells were harvested and extruded into liquid nitrogen. Aliquots were tested by western blot and the per-cell expression of both ORFs was indistinguishable from puromycin-selected material described previously (Fig. S1); transfection efficiency was assessed at >95% by indirect immunofluorescence of expressed ORF proteins. The median lysine and arginine heavy isotope incorporation levels for cell lines presented in this study were > 90%, determined as previously described (Taylor et al, 2013). RNase. These samples were reduced, alkylated and run until the dye front progressed ~6 mm on a 4-12% Bis-Tris NuPAGE gel (Life Technologies, as per manufacturers instructions). The gels were subsequently subjected to colloidal Coomassie blue staining (Candiano et al, 2004) and the sample regions ("gel-plugs") excised and processed for MS analyses, as described below.

Split-Tandem Affinity Capture
400 mg of light (L) and heavy (H) pLD401 transfected cell powders, respectively, were extracted and clarified as above. These extracts were subjected to affinity capture on 20 μl α-FLAG magnetic medium, 30 min at 4°C, followed by native elution with 50 μl 1 mg/ml 3xFLAG peptide (15 min, RT). 45 μl of the elution were subjected to subsequent affinity capture upon 20 μl α-ORF1 magnetic medium, resulting in a 45 μl supernatant (Sup) fraction depleted of ORF1p. Finally, the material was eluted (Elu) from the α-ORF1p medium in 45 μl 2.2x LDS sample loading buffer by heating at 70°C for 5 min with agitation. To enable quantitative comparisons of fractions the samples were combined, respectively, as follows: 28 μl each of the L Sup with H Elu, and 28 μl each of the L Elu with H Sup. These samples were then prepared as gel-plugs (as above) and processed for MS analyses, as described below.

Mass Spectrometry Sample Preparation and Data Acquisition
Gel plugs were excised, cut into 1 mm cubes, de-stained, and digested overnight with enough 3.1 ng/μl trypsin (Promega, Madison, WI, #V5280) in 25 mM ammonium bicarbonate to cover the pieces. In RNase-sensitivity and split-tandem SILAC analyses based on pLD401, as well as in vitro protein exchange experiments based on pMT302 and pMT289, an equal volume of 2.5 mg/ml POROS R2 20 µm beads (Life Technologies #1112906) in 5% v/v formic acid, 0.2% v/v TFA was added, and the mixture incubated on a shaker at 4°C for 24 hr. Digests were desalted on Stage Tips (Rappsilber et al, 2007), Scientific) acquiring data-dependent fragmentation spectra (either CID spectra alone, or CID and HCD spectra).

Mass Spectrometry Data Analysis
Raw files were submitted to MaxQuant (Cox & Mann, 2008) version 1.5.2.8 for protein identification and isotopic ratio calculation. Searches were performed against human protein sequences (UP000005640, April 2016), custom L1 ORF1p and ORF2p protein sequences, common exogenous contaminants, and a decoy database of reversed protein sequences. Search parameters included fixed modification: carbamidomethyl (C); variable modification: Arg10, Lys8, methionine oxidation; razor and unique peptides used for protein quantitation; requantify: enabled. Contaminants, low-scoring proteins and proteins with one razor+unique peptides were filtered out from the MaxQuant output file "proteingroups.txt". The list of contaminants was uploaded from the MaxQuant web-site (http://www.coxdocs.org/; "contaminants"). Additionally, proteins with the "POTENTIAL CONTAMINANT" column value "+" were filtered out. Proteins with at least 2 razor+unique peptides were retained for the analysis. H/(H+L) and L/(H+L) values were derived from unnormalized "ratio H/L" values and were used for plotting label-swapped RNasesensitivity and split-tandem data. Unnormalized "ratio H/L" values were used to calculate H/(H+L) in ORF2p catalytic mutant comparisons and in vitro exchange experiments. These values have been referred to as "affinities" within the Supplementary Materials.
Normalization and clustering procedures applied to data presented in the figures (Table S1) are detailed below and also in the Supplementary Methods.
To plot RNase-sensitivity affinity capture results (Fig. 1C), these data were normalized such that proteins that did not change upon treatment with RNases are centered at the origin.
The mean value and standard deviation were calculated using the distribution of distances from the origin. The distance threshold for p-value = 0.001 was calculated using the R programming language. A circle with radius equal to the threshold was plotted. Points with distances higher than the threshold were marked as black. To plot split-tandem affinity capture results (Fig. 1F), these data were normalized such that the ORF1p affinity was set to 1 and the distribution median was maintained. Probabilities associated with selected clusters were calculated based on the frequency distributions of 2-and 3-node clusters present in the data. To plot ENand RTmutant affinity capture results (Fig. 3B), the matrix of detected proteins for each experiment (ENand RT -) was filtered to retain only proteins detected in at least two replicate experiments. The difference between the affinity value of ORF2p and 0.5 value was calculated for each experiment. The affinities of each protein were shifted by the calculated difference. To determine the statistical significance of differentially co-purified proteins between ENor RTand WT, respectively, we used a 1sample t-test and applied Benjamini-Hochberg p-value correction. To determine the statistical significance of differentially co-purified proteins between ENand RTwe used an unpaired t-test and applied Benjamini-Hochberg p-value correction. To plot in vitro dynamics (Fig. 4B, C), only proteins which were identified at all time points were used. The cosine similarity method was used to calculate distances between proteins, and hierarchical clustering was used to visualize these distances. To integrate and plot the combined data ( Fig. 5), we calculated Euclidean and cosine distances for each I-DIRT-significant protein pair present in each experiment. Euclidean distances were rescaled to the range (0, 0.9).
Proteins not detected in any common experiments were assign a Euclidian distance of 1 after rescaling. The total distance between protein pairs was calculated as d = log((Rescaled Euclidean distance) * (cosine distance)). This distance was rescaled to the range (0, 1). Hierarchical clustering was used to visualize the calculated distances.

Gene Ontology (GO) Analysis
Genes corresponding to the proteins previously reported as significant by I-DIRT (Taylor et al, 2013) were tested for statistical overrepresentation using the default settings provided by http://www.panthnerdb.org (Mi et al, 2017(Mi et al, , 2013, searches were conducted using GO complete molecular function, biological process, and cellular compartment: all results are compiled in Table S4.

Acknolwedgements
This work was supported in part by National Institutes of Health (NIH) grants P41GM109824 (to M.P.R.), P41GM103314 (to B.T.C.), and P50GM107632 (to J.D.B), and by the 5-100 Russian Academic Excellence Program. The mass spectrometric analysis of proteins co-purified with chromatin associated ORF2p (Table S3) was conducted within the NYU School of Medicine Proteomics Resource Lab, which is partially supported by the Laura and Isaac Perlmutter Cancer Center Support Grant, NIH P30CA16087, and NIH 1S10OD010582. This paper is subject to the NIH Public Access Policy. The authors declare no conflicts of interest. Table Legends   Table 1. Putative L1 interactors: Through a series of affinity capture experiments using I-DIRT, we characterized a set of putative host-encoded L1 interactors (Taylor et al, 2013).  such that proteins that did not change upon treatment with RNases are centered at the origin. A cut-off of p = 10 -3 for RNase-sensitivity is indicated by a light gray circle; proteins that are RNase-sensitive with a statistical significance of p < 10 -3 are outside the circle. Proteins previously ranked significant by I-DIRT analysis ( Table 1) are labeled and displayed in blue or magenta (as indicated); black nodes were not found to be significant by I-DIRT but were labeled if found to be RNase-sensitive; gray, unlabeled nodes were not found to be significant by I-DIRT. (D) Split-tandem affinity capture: L1 complexes were affinity captured by ORF2p-3xFLAG. After native elution with 3xFLAG peptide, this fraction was subsequently depleted of ORF1p containing complexes using an α-ORF1 conjugated magnetic medium, resulting in a supernatant fraction depleted of ORF1p-containing complexes. The α-ORF1 bound material was then released with LDS, yielding an elution fraction enriched for ORF1p-containing complexes. Representative SDS-PAGE / Coomassie blue stained results for each fraction are shown. (E) SILAC duplicates, two supernatants and two elution, were cross-mixed to enable an assessment of the relative protein content of each fraction by quantitative MS. (F) The results from split-tandem affinity capture graphed as the fraction of each protein observed in the elution sample. In order to easily visualize the relative degree of co-partitioning of constituent proteins with ORF1p, these data were normalized, setting the fraction of ORF1p in the elution to 1. Proteins which were previously ranked significant by I-DIRT analysis are labeled and displayed in blue or magenta (as indicated); gray, unlabeled nodes were not found to be significant by I-DIRT.

Figure and
MOV10 is marked with a dagger because in one replicate of this experiment it was detected by a single unique peptide, whereas we have enforced a minimum of two peptides (see Methods) for all other proteins, throughout all other proteomic analyses presented here.

Figure 2. Immunofluorescent Imaging Reveals ORF1p is Required for Nuclear ORF2p:
Puromycin-selected HeLa-M2 cells containing pLD401 (Tet promoter, [ORFeus-Hs] full L1 coding sequence, ORF2p-3xFLAG, top two rows) or pLD561 (Tet promoter, ΔORF1, ORF2p-3xFLAG, bottom row) were plated on fibronectin-coated coverslips and induced for 24 hr with doxycycline prior to fixation and staining. With pLD401, the previously-observed pattern of cytoplasmic-only ORFs (top row) and a new pattern of pairs of cells expressing ORF2p in the nucleus (middle row) were apparent. When ORF1p is omitted from the construct (pLD561, bottom row), nuclear ORF2p was not apparent. Scale bars: 10 micrometers.   Table 1 are highlighted if present. Three clusters were observed (as indicated). (C) The cosine distance between the observed I-DIRT significant proteins was plotted along with time.      * * * * * * * * * ** ** ** ** ** ** ** ** ** ** ** ** ** ** B C T a n d e m R N A s e R T m u t a n t E N m u t a n t E x c h a n g e