Innate, translation-dependent silencing of an invasive transposon in Arabidopsis

Co-evolution between hosts’ and parasites’ genomes shapes diverse pathways of acquired immunity based on silencing small (s)RNAs. In plants, sRNAs cause heterochromatinization, sequence-degeneration and, ultimately, loss-of-autonomy of most transposable elements (TEs). Recognition of newly-invasive plant TEs, by contrast, involves an innate antiviral-like silencing response. To investigate this response’s activation, we studied the single-copy element EVADÉ (EVD), one of few representatives of the large Ty1/Copia family able to proliferate in Arabidopsis when epigenetically-reactivated. In Ty1/Copia-elements, a short subgenomic mRNA (shGAG) provides the necessary excess of structural GAG protein over the catalytic components encoded by the full-length genomic flGAG-POL. We show here that the predominant cytosolic distribution of shGAG strongly favors its translation over mostly-nuclear flGAG-POL, during which an unusually intense ribosomal stalling event coincides precisely with the starting-point of sRNA production exclusively on shGAG. mRNA breakage occurring at this starting-point yields unconventional 5’OH RNA fragments that evade RNA-quality-control and concomitantly likely stimulate RNA-DEPENDENT-RNA-POLYMERASE-6 (RDR6) to initiate sRNA production. This hitherto-unrecognized “translation-dependent silencing” (TdS) is independent of codon-usage or GC-content and is not observed on TE remnants populating the Arabidopsis genome, consistent with their poor association, if any, with polysomes. We propose that TdS forms a primal defense against de novo invasive TEs that underlies their associated sRNA patterns.


23
Transposable elements (TEs) colonize and threaten the integrity of virtually all 24 genomes (Huang et al, 2012). Chromosomal rearrangements caused by their highly-25 repetitive nature (Fedoroff, 2012) are usually circumvented by cytosine methylation 26 and/or histone-tail modifications at their loci-of-origin. The ensuing heterochromatic 27 DNA is not conducive to transcription by RNA Pol II, bringing TEs into an epigenetically 28 silent transcriptional state (Allshire & Madhani, 2018). This "transcriptional gene 29 silencing" (TGS) is observed at the majority of TE loci in plants, including the model 30 species Arabidopsis thaliana, and causes, over evolutionary times, accumulating 31 mutations resulting in mostly degenerated, non-autonomous entities (Vitte & 32 Bennetzen, 2006;Civáň et al, 2011). Nonetheless, the genome-invasiveness of these 33 remnants remains evident by their methyl cytosine-marked DNA, which is perpetuated 34 over generations by METHYL-TRANSFERASE 1 (MET1), among other factors. MET1 35 reproduces symmetrical methylation sites from mother-to daughter-strands during 36 DNA replication (Kankel et al, 2003) aided by the (hetero)chromatin remodeler 37 DEFICIENT IN DNA METHYLATION 1 (DDM1) (Saze et al, 2003;Zemach et al, 38 2013). 39 Loss of MET1 or DDM1 functions in Arabidopsis leads to genome-wide 40 demethylation, transcriptional reactivation of many TE remnants, and mobilization of 41 a small portion of intact, autonomous TEs (Mirouze et al, 2009;Tsukahara et al, 2010). 42 Their proliferation together with genome-wide deposition of aberrant epigenetic marks 43 likely explains why met1 and ddm1 mutants accumulate increasingly severe genetic 44 and phenotypic burdens over inbred generations (Vongs et al, 1993). However, such 45 secondary events can be avoided by backcrossing the first homozygous generation of 46 strategy, only EVD spawns detectable RDR6-dependent shGAG siRNAs (Oberlin et 250 al, 2017), prompting us to explore the basis for this difference. Polysome association, 251 independently of translation efficiency, is the most decisive prerequisite for any given 252 RNA to engage the translation machinery. For instance, many non-coding RNAs are 253 mostly nuclear (Khanduja et al, 2016), and aberrant (e.g. uncapped and/or poly(A) -) 254 mRNAs are actively degraded by RQC, both of which explain their general absence 255 from polysomes (Doma & Parker, 2007). We conducted genome-wide correlation 256 analyses between steady-state transcript accumulation, polysome association, and 257 siRNA levels of reactivated TEs in the ddm1 versus ddm1 rdr6 background by 258 calculating the ratio of polysome-associated versus total mRNA levels. The same 259 approach was applied to Arabidopsis protein-coding compared to non-coding RNAs 260 used as references (Oberlin et al, 2017). This analysis revealed two distinct TE 261 13 populations according to the levels of associated RDR6-dependent siRNAs. On the 262 one hand, approximately ¾ of ddm1 de-repressed TEs (530/674) display varying 263 degrees of polysome association, some within the range of protein-coding genes (Fig. 264 4A, quartiles 1-3). However, RDR6-dependent siRNA production does not accompany 265 their reactivation presumably because of their low expression levels ( overrepresented on polysomes (Oberlin et al, 2017) and is the major, if not unique 279 source of EVD-derived siRNAs (Fig.1, S1, Fig.3). This analysis suggests, therefore, 280 that translation is the step stimulated by splicing-coupled PCPA of shGAG, upon which 281 RDR6 is recruited specifically onto this mRNA isoform. 282 283

Splicing-coupled PCPA promotes selective translation and PTGS initiation from 284
shGAG-like mRNA isoforms 285 14 To test if differential translation due to splicing-coupled PCPA indeed underlies siRNA 286 production from shGAG as opposed to flGAG-POL, we used GFP-EVDint/ter-GUS, from 287 which the two EVD RNA isoforms and associated siRNA production/activity patterns 288 are recapitulated (Fig.2). Of the shGAG-like shGFP-and flGAG-POL-like flGFP-  mRNAs, only the former produced a detectable protein under the form of free GFP 290 ( Fig.S4C-D) despite accumulation of both mRNAs ( Fig.2A, Fig.S4A). Free GFP levels 291 were increased in the rdr6 background (Fig.S4C), coinciding with increased shGFP-292 but unchanged flGFP-GUS-mRNA levels (Fig.2B). The lack of detectable GFP-GUS 293 fusion protein -the expected product of flGFP-GUS-in either WT or rdr6 backgrounds 294 ( Fig.2D, Fig.S4C-D) was not due to intrinsically poor translatability. Indeed,  was the sole protein detected in independent lines undergoing RDR6-dependent 296 PTGS of 35S:GFP-GUS, a construct identical to 35S:GFP-EVDint/ter-GUS, save the 297 shGAG intron and PCPA signal ( Fig.5A-B). As expected, the GFP-GUS fusion protein 298 and GFP-GUS mRNA levels were strongly enhanced in the rdr6 versus WT 299 background ( Fig.5A-B). Yet, in contrast to GFP-EVDint/ter-GUS, from which siRNAs are 300 restricted to shGFP, the siRNAs from GFP-GUS encompassed both the GFP and GUS 301 sequences (Fig.5A). These results therefore indicate that splicing-coupled PCPA 302 promotes selective translation of, and PTGS initiation from, shGAG-like as opposed 303 to flGAG-POL-like mRNA isoforms. 304 305

Intron retention causes selective nuclear seclusion of flGAG-POL-like mRNAs 306
What mechanism linked to splicing-coupled PCPA might underpin the differential 307 translation of shGAG-like versus flGAG-POL-like mRNAs? Noteworthy, splicing 308 generally enhances mRNA nuclear export and translation (Valencia et al, 2008;309 Sørensen et al, 2017). Conversely, polyadenylated, unspliced mRNAs are retained in 310 15 the nucleus in Arabidopsis and only exported to the cytoplasm upon splicing (Jia et al, 311 2020). Moreover, 5' splice motifs and U1 snRNP binding promote chromatin tethering 312 of long non-coding RNAs in animal cells (Lee et al, 2015;Yin et al, 2020). We thus 313 tested if intron-retention might promote nuclear sequestration of the unspliced flGFP-314 GUS and flGAG-POL or if, conversely, splicing might favor export of shGFP and 315 shGAG to the cytoplasm, thereby selectively promoting their translation. We 316 performed nucleo-cytosolic fractionation (Fig.S5C) to analyze the relative distributions 317 of EVD-derived RNA isoforms produced in 3S:EVDwt or 35S:GFP-EVDint/ter-GUS 318 plants, using spliced/unspliced isoform-specific PCR amplification. Additionally, 319 unspliced isoforms were selectively analyzed using qPCR primer sets designed to 320 amplify sequences located near the 3' end of flGAG-POL or flGFP-GUS, and absent 321 from shGAG and shGFP (Fig.1A, Fig.2A). A similar approach was used to differentiate 322 the unspliced versus spliced ACTIN mRNA (Fig.S5D). Finally, the nuclear-only 323 snoRNA U5 (Fig.5C) was used as a control to assess the quality of nuclear 324 enrichments. To optimize accumulation of both types of RNA isoforms, the 325 experiments were all conducted in the PTGS-deficient rdr6 background. 326 The analysis revealed strikingly distinct nucleo-cytosolic distribution patterns for 327 the full-length versus short spliced mRNAs from both systems. Indeed, while the 328 spliced shGFP and shGAG were found predominantly in the cytosol (Fig.5C), flGAG-329 POL and flGFP-GUS were strongly enriched in nuclear fractions (Fig.5C, Fig.S5D). To 330 validate that nuclear unspliced full-length transcripts are bona fide poly(A) + mRNAs as 331 opposed to nascent transcripts or splicing intermediates, cDNA from the same RNA 332 samples was synthesized using exclusively oligo-dT to capture polyadenylated RNAs 333 only. This approach generated comparable results (Fig.5D), indicating that nuclear 334 full-length transcripts are properly terminated mRNAs. Corresponding results were 335 16 obtained in epi15 F11 plants displaying endogenous EVD reactivation ( Fig.S5E-F). 336 Collectively, these findings suggest that the unique splicing behavior of EVD -which 337 is recapitulated in GFP-EVDint/ter-GUS -not only allows production of the GAG-338 encoding shGAG subgenomic mRNA, but simultaneously promotes nuclear retention 339 of flGAG-POL. This is likely contributing to the disproportionate translation of shGAG 340 over flGAG-POL, although we do not exclude the involvement of other processes. 341 Under these premises, splicing-coupled PCPA likely predisposes shGAG, as opposed 342 to flGAG-POL, to one or several co-translational processes which, in turn, signal(s) 343 RDR6 recruitment. 344 345 Saturation of co-translational mRNA decay unlikely triggers shGAG siRNA 346 production 347 In plants and fungi, decapping coupled to 5'->3' exonucleolytic activity operated by 348 cytosolic XRN proteins regulate the intrinsic half-life of most actively translated 349 transcripts by degrading decapped mRNAs after the last translating ribosome 350 (Kastenmayer & Green, 2000;Hu et al, 2009;Pelechano et al, 2015;Yu et al, 2016). 351 Of the three Arabidopsis XRNs, XRN2 and XRN3 are nuclear, whereas XRN4 is 352 cytosolic and, hence, mediates co-translational mRNA decay (Gregory et al, 2008;353 Kurihara, 2017;Yu et al, 2016). Remarkably, transcripts undergoing improper 354 decapping and/or XRN4-mediated exonucleolysis constitute competing substrates for 355 RDR6 in Arabidopsis (Gazzani, 2004;Gy et al, 2007a;Gregory et al, 2008;Moreno et 356 al, 2013;Martínez de Alba et al, 2015) (Fig.6A). For instance, loss-of-RDR6 function 357 suppresses the lethality of decapping mutants by preventing production of undesirable 358 siRNAs from hundreds of endogenous mRNAs (Martínez de Alba et al, 2015). 359 Conversely, loss of XRN4 activity enhances RDR6-dependent PTGS (Gy et al, 2007a;360 Gregory et al, 2008;Moreno et al, 2013). These observations strongly suggest that 361 RDR6-dependent PTGS takes over co-translational mRNA decay when this process 362 becomes saturated by highly abundant and/or highly translated mRNAs. 363 Likewise, we reasoned that intense translation might overwhelm XRN4-mediated 364 co-translational decay of shGAG and thereby concurrently promote RDR6 action 365 ( Fig.6A). This would predict an accumulation of RNA degradation fragments (reflecting 366 XRN4 activity) coinciding with siRNA accumulation. PARE (parallel amplification of 367 RNA ends) and related methods map mostly XRN4 products associated with co-368 translational decay as well as non-translational RNA cleavage events, e.g. miRNA-369 mediated slicing of non-coding RNAs (Gregory et al, 2008;Schon et al, 2018). We 370 therefore conducted nanoPARE analyses, which capture both capped and uncapped 371 RNA fragments (Schon et al, 2018), in ddm1 vs WT Arabidopsis (Fig.S6A). 372 Simultaneously, mRNA-seq (i.e. SMART-seq2) was conducted on the same RNA to 373 monitor gene expression (Schon et al, 2018). Analysis of TAS1c, which undergoes 374 miR173-mediated slicing, confirmed that the ensuing 3' RNA cleavage fragment, a 375 common substrate of XRN4 (Schon et al, 2018), was readily detected in both 376 backgrounds, despite spawning vast amounts of RDR6-dependent siRNAs (Fig.6B). 377 Analyzing EVD upon its reactivation in ddm1 revealed a low level of RNA degradation 378 fragments spanning the entirety of EVD despite the siRNAs being exclusively derived 379 from shGAG. Had RNA degradation contributed to siRNA biogenesis, these species 380 would be expected to be distributed along the entirety of EVD, encompassing both 381 shGAG and flGAG-POL. Inspection of the housekeeping ACT2 locus revealed a 382 similar ORF-spanning degradation pattern, albeit at substantially higher levels (~10-383 folds), presumably reflecting the higher transcript abundance. However, ACT2 does 384 not spawn siRNAs (Fig. 6B). These observations therefore reveal no overt correlation 385 18 between abundance of RNA degradation products, siRNA production and/or polysome 386 association. 387 The above results did not formally exclude the possibility that at least some EVD-388 associated degradation products identified by nanoPARE might contribute to siRNA 389 biogenesis via competing RDR6 vs XRN4 activities. This would be genetically 390 diagnosed by an increased accumulation of shGAG siRNAs in xrn4 in contrast to their 391 loss in rdr6 (Gy et al, 2007a;Gregory et al, 2008). To test this idea without the potential 392 complication of EVD overexpression artificially saturating XRN4 activity in 35S: EVDwt,393 we introgressed the xrn4 null-mutation into epi15 at the early F8 inbred generation, 394 when PTGS of EVD is commonly initiated (Marí Ordóñez et al, 2013). As negative 395 controls, we used loss-of-function alleles of nuclear XRN2 and XRN3, which, by not 396 contributing to co-translational mRNA decay, should not influence siRNA production. 397 Finally, the rdr6 mutation was introgressed in parallel, to prevent shGAG siRNA 398 biogenesis. We analyzed two-to-three independent lineages with WT versus 399 homozygous mutant backgrounds isolated from segregating F2s. However, neither 400 xrn4 nor xrn2/xrn3 differed from the WT background with regard to EVD expression, 401 copy number, or shGAG siRNA levels ( Fig.6C-E, Fig.S6B-G). In contrast, EVD 402 expression and copy numbers were increased in rdr6, coinciding with reduced shGAG 403 siRNA levels ( Fig.6F-H). We conclude from these collective results that saturation of 404 XRN4-dependent co-translational mRNA decay (Fig.6A) is unlikely to underlie shGAG 405 siRNA production. siRNAs are, instead, abruptly spawned from the middle up to the 406 3' end of shGAG, as if their production coincided with a discrete co-translational event 407 ( Fig.6A). A similar rationale should apply to the discrete shGFP-centric siRNA pattern 408 spawned from GFP-EVDint/ter-GUS (Fig.2C). to ORF lengths, we compared them to those of actively translated Arabidopsis 427 mRNAs. To exclude artefacts from transcripts with low coverage, we restricted our 428 analysis to the most abundant mRNA isoforms with coverage available for more than 429 70% of ORFs, as described (Sabi & Tuller, 2015). We found that shGAG ranks among 430 the top 4.01 and 2.77 % (in WT and rdr6 backgrounds respectively) of Arabidopsis 431 transcripts displaying the most intense stalling events (Fig.7C). Remarkably, 432 overlaying siRNAs and codon coverage intensity revealed that the intense stalling 433 position coincides nearly exactly with the 5' starting point of the RDR6-dependent EVD 434 20 siRNA pattern (Fig.7D). Stalling is likely causal, not a consequence of RDR6 435 recruitment, because it also occurs in the rdr6 background (Fig.7A). 436 To explore further a possible link between discrete, intense ribosome stalling and 437 RDR6 recruitment, RFPs were conducted in the 35S:GFP-EVDint/ter-GUS background. 438 As seen above for shGAG versus flGAG-POL in the EVD context, the analysis 439 confirmed the vastly disproportional translation of shGFP versus flGFP-GUS 440 ( Fig.S8B). It also identified a major stalling site only in the shGFP ORF (whose 441 detection was enhanced in the rdr6 background) in which two prominently covered 442 and consecutive codons (pos. 235-236) accounted for ~40% of footprints (Fig.S8C). 443 Similarly to shGAG, shGFP ranked among the top 3.21 to 4.14% Arabidopsis 444 transcripts displaying the most intense stalling events (Fig.S8D). Furthermore, this 445 stalling site was located between major peaks of shGFP siRNAs, in this case, in both 446 the 5' and 3' directions within GFP-EVDint/ter-GUS (Fig.S8E). intermediates that, in turn, serve as RDR6 substrates. 476 477 Ribosome stalling correlates with production of 5'-hydroxy 3'-cleavage 478 fragments that possibly serve as RDR6 substrates 479 As described above, nanoPARE in ddm1 did not reveal any discrete RNA products 480 with 5' ends mapping consistently at, or near, the stalling site in shGAG. We also failed 481 to detect such products using classic 5' RACE (Llave et al, 2002). Noteworthy, this 482 technique relies on a 5' monophosphate (5'P) for RNA ligation of 5' adaptors (Silber 483 22 et al, 1972;Wang & Fang, 2015). Intriguingly, 5'P was reported to be absent from 484 various 3' cleavage RNA fragments produced co-translationally in budding yeast, 485 including upon ribosome stalling (Peach et al, 2015;Navickas et al, 2020). A lack of 486 5'P is also strongly suspected for the 3' cleavage products of ribothrypsis (Ibrahim et 487 al, 2018). Since siRNA production from EVD initiates just downstream of the major 488 stalling site (codons 148-149), we thus considered the possibility that discrete To explore such a connection and simultaneously characterize and map the 494 5' ends of putative shGAG 3' cleavage fragments, we used the RtcB RNA ligase. RtcB 495 contributes to tRNAs splicing by ligating RNAs with 3'P ends (or 2',3'-cyclic phosphate) 496 to 5'OH ends, and was used previously to map co-translational RNA cleavage 497 fragments in yeast (Desai & Raines, 2012;Peach et al, 2015). A 5' RNA adaptor with 498 a 3'P end was therefore RtcB-ligated to total RNA extracted from plants expressing 499 35S:EVD or non-transgenic controls, both in the rdr6 background. Use of rdr6 500 prevented conversion of potential RDR6 templates into dsRNA as well as the 501 accumulation of confounding cleavage fragments potentially caused by the ensuing 502 secondary siRNAs. The ligated RNA was then subjected to reverse transcription using 503 EVD-specific primers surrounding the major stalling site ( Fig. 7F; Region #1), amplified 504 through PCR, and cloned following standard RACE procedures. Based on the EVD 505 ribosome footprint profile (Fig. 7B), we also investigated two additional regions more 506 covered with ribosomes than expected ( Fig. 7F; Regions #2 & 3). Only region #1 507 yielded detectable amplification products within the expected size range. Nonetheless, 508 23 gel excision within the anticipated size ranges followed by cloning was performed for 509 all regions in all genotypes (Fig. S10A-C). Sanger sequencing revealed that 30 out of 510 36 fragments cloned from region #1 displayed 5'OH ends consistently mapping at 511 nucleotides 447-448, strikingly defining the intense ribosome stalling site on shGAG 512 ( Fig. 7F, S10D) from which siRNA production is initiated (Fig. 7D). By contrast, the 513 clones obtained from regions #2 and #3 were either devoid of EVD sequences or 514 empty. These results are consistent with the notion that the intense ribosome stalling 515 event correlates with breakage of the shGAG RNA, and that the ensuing 5'OH 516 fragments serve as templates for RDR6 to initiate dsRNA production and downstream 517 siRNA processing. Given that XRNs require a 5-P for their 5'->3' exonucleolytic 518 activities (Stevens, 2001;Schon et al, 2018), this could explain the insensitivity of 519 shGAG-derived siRNA accumulation to any xrn mutation and to xrn4 in particular 520

Translation as an initiator of PTGS and epigenetic silencing 535
Protein synthesis is commonly merely seen as a target of PTGS by reducing the 536 amount of available RNA and/or interfering with translation. Our study identifies 537 translation also as a trigger of PTGS. This became evident after epigenetic reactivation 538 of EVD, from which splicing-coupled PCPA generates separate RNA isoforms from a 539 single transcription unit. Of the two, the shorter subgenomic shGAG RNA undergoes 540 disproportionate translation over flGAG-POL as an indispensable feature of Ty1/Copia 541 biology because this likely provides the stochiometric protein balance necessary for 542 efficient amplification and mobilization of the element. This process, however, 543 concomitantly stimulates RDR6 activity. shGAG translation efficacy per se is within 544 the range of moderately translated Arabidopsis mRNAs and is unlikely to explain this 545 effect, nor do GAG expression or abundance. Rather, an exceptionally intense and 546 highly discrete ribosome stalling event predisposes shGAG to RDR6-dependent 547 PTGS. Our data also suggest how intron-retention in combination with active splicing 548 accounts for the mostly nuclear versus cytosolic localization of flGAG-POL versus 549 shGAG, respectively. Their asymmetrical subcellular distribution concurrently 550 rationalizes (i) the disproportionate translation efficacies of each mRNA, (ii) the 551 shGAG-centric distribution of translation-dependent EVD-derived siRNAs and, 552 consequently, (iii) the contrasted sensitivity of each isoform to cytosolic PTGS. 553 Splicing-coupled PCPA probably underlies most, if not all, of features i-iii because they 554 were recapitulated with the GFP-EVDint/ter-GUS construct containing the shGAG intron 555 and proximal PCPA signal (Fig.2, S4 and 5). Since splicing-coupled PCPA is at the 556 very core of the Ty1/Copia genome expression strategy (Oberlin et al, 2017), the 557 process described here for EVD is likely to be broadly applicable. The vast majority of ddm1-reactivated TEs that spawn RDR6-dependent siRNAs is 584 composed of LTR/Gypsy elements (Fig.4), which is the family most prominently 585 associated with easiRNA production (Creasey et al, 2014;Borges et al, 2018). 586 Arabidopsis LTR/Gypsy elements generally display significantly shorter-than-full- Capped but poly(A) -5' RNA fragments are expected to be concurrently 698 produced during the RNA breakage process, that should also be amenable to RDR6 699 activity. Yet, only a small siRNA peak is detected just upstream of the intense stalling 700 site on shGAG. Unlike the XRN4-protected 5'OH 3' fragments, however, these 701 5' fragments would be a priori devoid of features preventing their SKI2-mediated 702 exosomal 3'-5' degradation, which indeed can outcompete RDR6 action (Zhang et al, 703 2015;Branscheid et al, 2015). Differential competition between SKI2-mediated 704 degradation versus RDR6 activity, as previously reported for some miRNA targets 705 (Branscheid et al, 2015), might explain why the siRNAs pattern from shGFP spans 706 31 both the 5' and 3' sequences surrounding the identified stalling site whereas these are 707 only located 3' to the shGAG stalling site of (Fig.7D, S8E). In principle, RDR6 could 708 also pick up a multitude of RNA cleavage fragments predictably produced via siRNA-709 guided cleavage of shGAG by AGO1/AGO2-RISCs. However, RISC-mediated slicing 710 produces 5'P termini (Martinez & Tuschl, 2004), which would qualify these RNAs as 711 XRN4-, as opposed to RDR6-, substrates, therefore unlikely contributing prominently 712 to shGAG siRNA production. A final, outstanding aspect of TdS pertains to the 713 mechanism whereby 5'OH fragments are generated. In budding yeast, the metal-714 independent endonuclease Cue2 was recently shown to cleave, within the colliding 715 ribosome's A site, mRNAs undergoing stalling-induced no-go decay, which generates procedures. However, while its initiation strongly resembles that of mammalian 732 ribothrypsis, TdS is unlikely to be ubiquitous in plants, since the aforementioned RNA 733 products, by directly engaging RDR6 for amplified siRNA production, would promote 734 degradation of the entire mRNA pool independently of its stalled or even merely 735 translated status. While this would be highly detrimental as a common form of 736 endogenous gene regulation, the process seems particularly well-suited to eliminate 737 highly proliferating foreign RNAs such as those of viruses and TEs. We suspect that 738 TdS might also represent a yet unexplored trigger for PTGS of transgenes encoding, 739 in particular, non-plant ORFs with suboptimal translation features. 740 Author contributions:    shGFP term. GFP-GUS term.       High and low molecular-weight RNA analysis using a GFP or GUS probe in two independent transgenic lines from each construct in the WT or rdr6 background. mRNA isoforms are indicated with arrows and correspond to the transcripts depicted in Fig.2A. EtBr staining of the agarose gel and miR171 probe serve as loading control for mRNAs and sRNAs, respectively. (B) Western analysis of the translation products from GFP and GFP-GUS transcripts. Coomassie (coom.) staining as a loading control. (C) Nuleo-cytosolic distribution of 35S:EVD and 35S:GFP-EVDint/ter-GUS RNA isoforms in rdr6 relative to that of ACT2 analyzed by qPCR. RNA extracted from Total, nuclear (Nucl) and cytoplasmic (Cyto) fractions was reverse transcribed with random hexamers and oligo(dT). snoRNA U5 is shown as a nuclear-only RNA control. (D) Same as in (C) but using exclusively oligo(dT) to reverse transcribe poly(A)+ RNAs. Both in (C) and (D), qPCR was performed on n=3 biological replicates; bars: standard error. (*) = p-value < 0.05, (**) = p-value < 0.01 (two-sided t-test between indicated samples). During co-translational decay, mRNA turnover is initiated by decapping of actively translated mRNAs. This exposes 5' monophosphate (5'PO4) groups required for the 5'->3' exonucleolytic activity of XRNs. As evidenced by the RDR6-dependent production of siRNAs from dozens of endogenous loci in decapping and xrn4 mutants of Arabidopsis, plant mRNAs engaged in co-translational decay can become substrates for siRNA biogenesis if they are not degraded by, or their levels saturate, this process. While depicted here, for simplicity, as a co-translational event, RDR6 action onto such RNA likely occurs in specialized "siRNA bodies" adjacent to P-bodies (B) EVD, ACT2 and TAS1c capped and uncapped 5'ends from nanoPARE and Smart-seq2 libraries along 20-21 nt siRNA in WT and ddm1. (C-E) EVD genomic proliferation in homozygous xrn4 mutant and WT backgrounds in F2 plants from a cross between xrn4 and epi15 exhibiting active EVD mobilization. (C) sRNA blot analysis using an anti-GAG probe. (D) Relative expression levels of shGAG and GAGPOL normalized to ACT2 and to AT4G26410 levels. (E) EVD genomic copy number quanti cation by qPCR. qPCR analysis was performed on three biological replicates for controls and the three independent WT and mutant F2 lines displayed in B. Lack of 5'PO prevents XRN 5'->3' exonucleolytic activity (see Fig.6A), granting the RNA to be used as template by RDR6.

GFP-EVD int/ter -GUS
(F) Overlap between ribosomal footprints and mapping of 5'OH ends from 35S:EVD in rdr6 cloned through RtbC ligation. Regions investigated are highlighted in grey. 5'OH ends were only successfully cloned from region #1. Alignment of sequenced clones to EVD is displayed in Fig.S9.