Translation is required for miRNA-dependent decay of endogenous transcripts

Posttranscriptional repression by microRNA (miRNA) occurs through transcript destabilization or translation inhibition. Whereas RNA degradation explains most miRNA-dependent repression, transcript decay occurs co-translationally, raising questions regarding the requirement of target translation to miRNA-dependent transcript destabilization. To assess the contribution of translation to miRNA-mediated RNA destabilization, we decoupled these two molecular processes by dissecting the impact of miRNA loss of function on cytosolic long noncoding RNAs (lncRNAs). We show, that despite interacting with miRNA loaded RNA-induced silencing complex (miRISC), the steady state abundance and degradation rates of these endogenously expressed non-translated transcripts are minimally impacted by miRNA loss. To validate the requirement of translation for miRNA-dependent decay, we fused a miRISC bound lncRNA, whose levels are unaffected by miRNAs, to the 3’end of a protein-coding gene reporter and show that this results in its miRNA-dependent transcript destabilization. Furthermore, analysis of the few lncRNAs whose levels are regulated by miRNAs revealed these tend to associate with translating ribosomes and are likely misannotated micropeptides, further substantiating the necessity of target translation for miRNA-dependent transcript decay. Our analyses reveal the strict requirement of translation for miRNA-dependent transcript destabilization and demonstrate that the levels of coding and noncoding transcripts are differently affected by miRNAs.

expressed non-translated transcripts are minimally impacted by miRNA loss. To 23 validate the requirement of translation for miRNA-dependent decay, we fused a 24 miRISC bound lncRNA, whose levels are unaffected by miRNAs, to the 3'end of a 25 protein-coding gene reporter and show that this results in its miRNA-dependent 26 transcript destabilization. Furthermore, analysis of the few lncRNAs whose levels are 27 regulated by miRNAs revealed these tend to associate with translating ribosomes and 28 are likely misannotated micropeptides, further substantiating the necessity of target 29 translation for miRNA-dependent transcript decay. Our analyses reveal the strict 30 requirement of translation for miRNA-dependent transcript destabilization and 31 demonstrate that the levels of coding and noncoding transcripts are differently affected 32 by miRNAs. Post-transcriptional regulation of gene expression by microRNAs (miRNAs) is 3 widespread in eukaryotes and impacts diverse biological processes in health and 4 disease [1,2]. Most mature miRNAs are the product of a relatively complex biogenesis 5 process. Primary miRNA transcripts, that generally depend on RNA Polymerase II for 6 transcription, are initially processed by the nuclear enzyme DROSHA and its cofactor 7 DGCR8 into a premature hairpin RNA of ~60 nucleotides in length (pre-miRNA 8 transcript) [3]. Pre-miRNAs are exported into the cytoplasm where they undergo a 9 second round of processing by DICER resulting in a ~22 nucleotide long double-10 stranded RNA duplex [4]. Loss of function mutations in any of the miRNA processing 11 factors result in complete depletion of most miRNA species [5]. Argonaute proteins 12 (AGO) bind mature miRNAs and guide target recognition of the RNA-inducing 13 silencing complex (RISC). In mammals, target recognition relies primarily on 14 complementarity between the miRNA seed region (position 2-8 of the mature miRNA) 15 and miRNA recognition elements (MREs) in the target [6]. 16

Posttranscriptional repression by miRNAs occurs by translation inhibition or transcript 17
decay [2]. The contributions of RNA destabilization and translation inhibition to miRNA 18 repression have been extensively studied [7,8]. These studies support the general 19 consensus that, translation inhibition precedes transcript deadenylation and decay [9-20 11], which in turn, is thought to account for most miRNA-dependent repression [9, 10, 21 12]. The coupling between translation inhibition and transcript destabilisation is further 22 substantiated by evidence that protein-coding transcripts undergoing miRNA-23 dependent repression associate with translating ribosomes [13][14][15][16][17][18][19], and that most 24 miRNAs loaded into RISC (miRISC) co-localize with polysomes [20][21][22]. 25 These observations have raised questions regarding the requirement of translation for 26 miRNA-dependent transcript decay. A number of experiments relying on the analysis 27 of reporter constructs, revealed that transcript decay occurs even when translation 28 initiation or elongation are impaired [23][24][25]. However, it is hard to reconcile the extent 29 of target repression reported in these studies (up to five-fold) with the well-established 30 impact of most miRNAs on endogenous transcript abundance, which rarely exceeds 31 2-fold [9,10]. This has prompted concerns on whether exogenously expressed 32 reporters faithfully recall the behaviour of most endogenously expressed transcripts. 33 To assess the requirement of translation for RNA destabilization of endogenous 1 miRNA-targets and to overcome some of the limitations that may arise from using 2 exogenous reporters, we took advantage of endogenously expressed cytosolic 3 intergenic long noncoding RNAs, lncRNAs. This class of noncoding transcripts rarely 4 associate with ribosomes [26] and have been previously shown to interact with miRISC 5 machinery [27]. These transcripts thus provide a unique opportunity to address the 6 outstanding question of whether miRNA-dependent decay occurs in the absence of 7 translation. Specifically, we used 4-thio-uridine (4sU) to assess genome wide decay 8 rates in wild-type (WT) and miRNA depleted cells. Our genome-wide analysis revealed 9 that the decay rates of protein coding miRNA targets are significantly reduced upon 10 miRNA loss whereas those of lncRNAs are only minimally impacted. Putative 11 micropeptides were enriched among lncRNAs responsive to changes in miRNA 12 abundance suggesting that translation is required for miRNA-dependent decay. We 13 validated this hypothesis experimentally by inducing association of candidate lncRNA 14 with translating ribosomes and found that this is sufficient to induce miRNA-dependent 15 decay, further substantiating the prerequisite of translation for miRNA-dependent 16 transcript decay.  Figure S1B). Since lncRNAs are in general more lowly 26 expressed than mRNAs, the proportion of lncRNAs bound by AGO2 may be higher 27 than what is detected. The fraction of cytosolic lncRNAs bound by AGO2 with (6%) 28 and without (7%) experimental evidence of ribosomal association is statistically 29 indistinguishable (two-tailed Fisher's exact test p=0.8), suggesting that AGO2 binding 30 is independent of translation. We conclude, that consistent with previous analysis, 31 most cytosolic lncRNAs do not stably associate with translating ribosomes [26], but 32 are nevertheless targeted by miRISC [27], and are therefore, uniquely suitable to 33 assess the impact of miRNAs on endogenous transcript destabilization in absence of 1 translation. 2 3

Steady-state expression of noncoding transcripts is minimally impacted by 4 miRNAs 5
We first sought to determine whether cytosolic lncRNA expression was post-6 transcriptionally regulated by miRNAs. We took advantage of a mESC cell line 7 containing two Cre/LoxP sites flanking the Dicer RNAse III domain on exon 21, and a 8 Cre recombinase gene expressed under the control of a 4-hydroxytamoxifen(4-OHT)-9 inducible promoter [31,32]. Exposure of these cells to 4-OHT leads to LoxP site 10 recombination and strong depletion of DICER (Supplementary Figure S1C). 11 Conditional loss of DICER function minimally impacts cell proliferation (Supplementary 12 Figure S1D) and the transcript and protein levels of (Supplementary Figure  We profiled small RNA expression following DICER loss of function and found that 8 20 days after 4-OHT addition, mature miRNA levels are reduced by ~80% ( Figure 1C). 21 We validated these results, by RT-qPCR, for miR-290 and miR-295, which are among 22 the most abundant miRNAs in mESCs [34] (Supplementary Figure S1H). Decreased 23 levels of these miRNAs is associated, as expected, with a significant increase in the 24 levels of some of their well-established targets [35] (Supplementary Figure S1I). 25 To assess the genome-wide impact of miRNA loss on mRNA and lncRNA expression, 26 we used data from our previously published transcriptome-wide expression profiling 27 following loss of DICER experiment in these cells [28]. As expected, and consistent 28 with the role of miRNAs on posttranscriptional repression of protein-coding gene 29 expression, we found that mRNA levels increased moderately but significantly 30 following Dicer loss of function ( Figure 1D). The fold-increase in expression, relative 31 to control, in miRNA depleted mESCs is significantly higher (two-tailed Mann-Whitney 32 test, p<1.4X10 -10 ) for transcripts with experimental evidence for AGO2 binding ( Figure  33 1E), supporting that the observed changes in mRNA expression are, at least in part, 34 a consequence of mRNA alleviation from miRNA-mediated repression. In contrast to 1 mRNAs, we found that lncRNA expression was minimally impacted by miRNA 2 depletion ( Figure 1F). Specifically, and in contrast to mRNAs, lncRNA steady-state 3 abundance is slightly decreased in miRNA depleted cells ( Figure 1F). This small 4 decrease is likely an indirect effect of miRNA loss. Specifically, decreased levels of 5 miRNAs are expected to result in increased steady state abundance of targets as 6 observed for mRNAs ( Figure 1F), whereas the impact of miRNA depletion is similar 7 for both subcellular classes of lncRNA independent of co-localization with miRISC 8 ( Figure 1F). We conclude that, despite interacting with miRISC, cytosolic lncRNA 9 transcript levels are not directly controlled by miRNAs ( Figure 1F). 10

11
No evidence for miRNA-dependent destabilization of noncoding transcripts 12 Steady-state transcript abundance depends on the rates of transcription, processing 13 and degradation but only the degradation is directly controlled by miRNAs. To 14 determine transcriptome-wide differences in degradation rate between miRNA 15 depleted and control mESCs we performed, in duplicate, 4-thio-uridine (4sU, 200uM) 16 metabolic labelling of RNA for 10 and 15 minutes, on mESCs 8 days after induction of 17 DICER loss of function and control mESC. We sequenced total RNA and quantified 18 intron and exon expression transcriptome wide from the pre-existing and newly 19 synthetized RNA fractions. (Figure 2A, Methods). Principal component analysis of the 20 gene expression estimates across the different samples revealed that the RNA 21 fraction is the strongest discriminator between estimates followed by miRNA content 22 and lastly by biological replicate . Degradation rates, 23 that we estimated using INSPEcT ( [36], methods) for the two different pulses durations 24 (10 and 15 minutes) are highly correlated for both cell types (R 2 >0.75, Figure 2B-C). 25 We used an alternative method (transcription block by Actinomycin-D) to validate the 26 estimated differences in transcript stability between wild-type and miRNA depleted 27 cells for a subset of transcripts spanning a range of fold-differences in degradation 28 rates (Pearson R 2 =0.58, Supplementary Figure 2D). 29 Next, we identified genes whose degradation rate is significantly different between 30 miRNA-depleted and control mESCs (10 and 15 minute pulse, Figure 2D and 31 Supplementary Figure S2E, respectively) and found that as expected, mRNAs are 32 significantly more often stabilized in miRNA depleted mESCs relative to control. 33 Finally, and consistent with a role of miRNA in controlling the observed differences in 34 degradation rates, transcripts whose decay rates are significantly decreased, following 1 miRNA depletion, have a significantly higher density of miRISC clusters (10 and 15 2 minute pulse, Figure 2E and Supplementary Figure S2F, respectively). 3 In contrast with mRNAs and in line with the observed changes in steady state 4 abundances, we found that the degradation rates of cytosolic lncRNAs are minimally 5 impacted by miRNA depletion, with only a few displaying significant differences in 6 degradation rate (10 and 15 minutes pulse, Figure 2D and Supplementary Figure 2E). 7 Specifically, most cytosolic lncRNAs behave similarly to nuclear lncRNAs (10 and 15 8 minutes pulse, Figure 2F and Supplementary Figure 2G, respectively). The decrease 9 in degradation rate between wild-type and miRNA depleted cells is likely to be, at least 10 in part a consequence of well described compensation mechanisms [37-39] to account 11 for decreased synthesis rates between the two cell types (Supplementary Figure  12   S2H). The analysis of steady-state abundance and degradation rates following loss of 13 Dicer function indicate that, in contrast with coding transcripts, cytosolic lncRNAs are 14 resilient to miRNA-mediated destabilization. 15 16

Micropeptide encoding transcripts undergo miRNA dependent destabilization 17
Next, since our analysis of ribosomal profiling data indicated that a small fraction of 18 cytosolic lncRNAs is ribosome-bound ( Figure 1A), we investigated whether 19 association with translating ribosomes would contribute to the impact of miRNAs on 20 the degradation rates of some cytosolic lncRNAs. As expected, mRNAs are 21 significantly more efficiently translated than lncRNAs but interestingly, the translation 22 efficiency of cytosolic lncRNAs, as a class, is significantly higher than that of nuclear 23 lncRNAs indicating that some might encode micropeptides ( Figure 3A) Table S1). These transcripts are almost 3 times more likely to be 30 bound by ribosomes than are other cytosolic lncRNAs ( Figure 3B) and their translation 31 efficiency is significantly higher than that of cytosolic lncRNAs (p<6X10 -5 , two-tailed 32 Mann-Whitney U test, Figure 3C ) and more similar to that of mRNAs (p<1X10 -4 , two-33 tailed Mann-Whitney U test, Figure 3C) consistent with some of these transcripts 34 encoding micropeptides. We separated micropeptides from bona fide cytosolic 1 lncRNAs and found that fold change in degradation rate of micropeptides in miRNA 2 depleted cells relative to control, is similar to what is obtained for mRNAs and 3 significantly different from what is observed for bonafide lncRNAs ( Figure 3D) 4 indicating further the requirement of translation for miRNA-dependent transcript 5 destabilization. 6 7 miRNA impact coding but not noncoding transcript stability 8 Our transcriptome wide analysis indicates that translation is required for miRNA 9 dependent target destabilization. To test this hypothesis, we selected one cytosolic  lncRNA-c1) and transfected this construct into wild-type and miRNA depleted mESCs 27 (8 days after induction of Dicer loss of function). As controls, we transfected GFP and 28 lncRNA-c1 expressing constructs. As expected, the expression of lncRNA-c1 and 29 GFP is more similar between wild-type and miRNA-depleted cells than is the 30 expression of GFP-lncRNA-c1, whose levels significantly increase in miRNA depleted 31 cells (paired two-tailed t-test p-value < 0.02, Figure 4B), consistent with its miRNA 32 dependent destabilization in wild-type cells. If association with the translation 33 machinery is sufficient to induce miRNA-dependent decay of a miRISC-bound 34 noncoding transcript, one would expect introduction of a missense mutation in a 1 protein-coding miRNA target to decrease its miRNA-induced decay. Indeed, 2 introduction of a missense mutation disrupting the Cdkn1a start codon 3 (Supplementary Figure S4A-C) significantly decreases mutant Cdkn1aDATG levels in 4 miRNA depleted mESCs (paired one-tailed t-test p-value < 0.05, Figure 4C). 5 Given that all constructs are under the control of the same promoter (T7), this increase 6 is likely a consequence of increased stability, as confirmed by qPCR analysis following 7 8h of transcription inhibition through actinomycin-D treatment (paired two-tailed t-test 8 p-value < 0.05, Supplementary Figure S4D). 9 LncRNA-c1 is a predicted target of the miR-290/295 family (Supplementary Figure  10   S4E). To validate that these miRNAs are indeed contributing to miRNA-dependent 11 repression of GFP-lncRNA-c1, we co-transfected mESCs with GFP-lncRNA-c1 12 expressing vector and miR-294-inhibitors. We note a significantly higher expression 13 of GFP-lncRNA-c1 in the inhibitor transfected cells compared to cells transfected with 14 negative control (unpaired two t-test p-value < 0.001, Figure 4D). We used site-15 directed mutagenesis to mutate three miRNA recognition elements (MREs) for highly 16 expressed miRNAs within GFP-lncRNA-c1 (hereafter GFP-lncRNA-c1ΔMRE). As 17 expected, reintroduction of miRNA mimics in DICER depleted mESC impacts the 18 levels of wild-type GFP-lncRNA-c1 more than it does levels of GFP-lncRNA-c1ΔMRE 19 (paired one-tailed t-test p-value<0.05, Supplementary Figure S4F). The levels of 20 GFP-lncRNA-c1 in wild-type mESC is also significantly lower than the level of GFP-21 lncRNA-c1ΔMRE (paired t-test p-value<0.05, Supplementary Figure S4G). These 22 results are consistent with these MREs' contribution to wild-type GFP-lncRNA-c1 23 miRNA-dependent repression. Therefore, and as expected, the relative increase of 24 GFP-lncRNA-c1 levels in miRNA depleted mESCs relative to wild-type mESC is 25 significantly higher than the increase in levels of GFP-lncRNA-c1ΔMRE (paired two 26 tailed t-test p-value < 0.05, Figure 4E). The presence of MRE for other mESC 27 expressed miRNA (Supplementary Table S2) is likely to explain why mutation of 28 miR290/295 MRE alone is not sufficient to entirely block miRNA-dependent GFP-29 lncRNA-c1 destabilization. 30 We conclude that association with translating ribosomes is required for miRNA-31 dependent transcript destabilization and that noncoding transcripts are bound but not 32 post-transcriptionally regulated by miRNAs. 33 CONCLUSION 1 2 Posttranscriptional regulation by miRNAs leads to translational inhibition or transcript 3 destabilization [6]. Whereas the general consensus is that most miRNA-induced 4 changes can be explained by transcript destabilization [9, 10], increasing evidence 5 suggests that miRNA-dependent mRNA decay occurs co-translationally [13][14][15][16][17][18][19][20][21][22], 6 raising questions about the ability of miRNAs to posttranscriptionally regulate the 7 levels of noncoding transcripts. 8 Supporting different outcomes upon miRISC binding to coding and noncoding 9 transcripts, is recent evidence that these two classes of transcripts have distinct 10 interaction dynamics with processing bodies (PB) [42], the subcellular compartment 11 where miRNA-dependent destabilization is thought to occur [43]. Specifically, and in 12 contrast with miRNA-bound mRNAs, which localise to the core of PB, miRNA-bound 13 lncRNAs interact transiently and tend to locate to the PB periphery, a pattern that might Our transcriptome wide analysis following miRNA loss revealed, that in contrast with 29 mRNA, cytosolic lncRNA's steady state abundance significantly decreases in miRNA 30 depleted cells, suggesting this class of transcripts is not efficiently posttranscriptionally 31 regulated by miRNAs. To assess the direct impact of miRNA regulation on cytosolic 32 lncRNAs, we investigated, using RNA metabolic labelling, differences in the 33 degradation rates of these transcripts in wild-type and miRNA-depleted cells. This 34 analysis revealed that cytosolic lncRNAs degradation rates decrease less than the 1 degradation rates of mRNAs and to a similar extent as the degradation rates of nuclear 2 lncRNAs, that are not expected to be regulated by miRNAs. The decrease of lncRNA 3 degradation rates in miRNA depleted cells is likely the result of coupling between RNA 4 synthesis and decay which has been proposed as a mechanism to ensure gene 5 expression homeostasis [37][38][39]. While the decrease in degradation rates is a general 6 phenomenon in miRNA-depleted mESCs (Supplementary Figure 2H), the increased 7 stabilization of coding transcripts in near-absence of miRNA is likely to obscure such 8 effects for mRNAs. 9 Finally, we show that the stabilities of putative micropetides and mRNAs are similarly 10 impacted by miRNAs, further supporting the requirement of translation for miRNA 11 dependent regulation of endogenously expressed transcripts. 12 To validate this hypothesis, we selected one cytosolic lncRNA, bound by AGO2 and 13 with functional binding sites for miR-290/5 family, and forced its association to More generally the present results also imply that miRISC binding, per se, is not 4 sufficient to determine the outcome of bound targets suggesting the requirement of 5 further yet unidentified molecular partners. 6 In summary, the analysis of endogenously expressed and miRISC bound noncoding 7 transcripts provides further evidence that translation is indispensable for miRNA-8 dependent regulation of endogenous transcripts, suggesting the requirement of further 9 molecular partners and highlighting differences in posttranscriptional regulation of 10 coding and noncoding RNAs.  1 Mouse embryonic stem cell culture 2 Feeder depleted mouse DTCM23/49 XY embryonic stem cells [28,32,54] were grown 1 on 0.1% gelatin-coated tissue culture treated plates in a humidified incubator with 5% 2 (v/v) CO2 at 37ºC in 1X DMEM medium supplemented with 1x Non-Essential Amino 3 Acids, 50 uM β-mercaptoethanol, 15% Fetal Bovine Serum, 1% 4 Penicillin/Streptomycin, and 0.01% of Recombinant mouse Leukemia Inhibitory factor. 5 Cultures were maintained by passaging cells every 48 hours (replating density ~ 6 3.8*10 4 cells/cm 2 ). Unless stated otherwise, to induce loss of Dicer function, cells were 7 cultured in mESC growth media supplemented with 800 nM tamoxifen previously 8 (H+L)) in 5% skim milk in TBS-for 1h at room-temperature. Immunoblots were 26 developed using the WesternBright ECL premixed Peroxide and ECL solutions and 27 detected using an imaging system (Vilber Fusion Chemiluminescence). 28 Following detection, secondary antibody coupled with the HRP was deactivated by 29 washing the membrane two times for 20 minutes with 1% (w/v) Sodium-Azide in TBS-30 T and the membrane incubated two hours at 4 ºC with 5% (w/v) skimmed milk in TBST 31 containing primary antibody for the ACTIN-β loading control (1:4000). The membranes 1 were washed three times for 15 minutes in fresh TBS-T and incubated for one hour at 2 room temperature with the secondary antibody coupled with horseradish peroxidase 3 in 5% skimmed milk in TBS-T (for ACTIN-β= 1:4000 Goat Anti-mouse IgG). Washing 4 and protein detection were performed as previously described. Five million DTCM23/49 XY mESCs (WT and miRNA depleted) were seeded and 28 allowed to grow to 70-80% confluency (approximately 1 day). 4sU was added to the 29 growth medium (final concentration of 200 µM) and cells were incubated at 37 ºC for 30 10 or 15 minutes. RNA was extracted using Trizol, according to manufacturer 31 instructions and DNAse treated using RNeasy on column digestion according to 1 manufacturer's instructions. 100 µg of RNA was incubated for 2 h at room temperature 2 with rotation in 1/10 volume of 10X biotinylation buffer (Tris-HCl pH 7.4, 10 mM EDTA) 3 and 2/10 volume of biotin-HPDP (1mg/ml in Dimethylformamide). Following 4 biotinylation, total RNA was purified through phenol:chloroform:isoamyl alcohol 5 extraction and precipitated with equal volume of Isopropanol and 1/10 volume of 5M 6 NaCl. RNA was washed once with 75% Ethanol and resuspended in in DEPC-treated 7 H2O. Equal volume of biotinylated RNA and pre-washed Dynabeads TM MyOne TM 8 Streptavidin T1 beads were mixed and incubated at room temperature for 15 minutes 9 under rotation. The beads were then separated using a DynaMag TM -2 Magnetic stand. 10 The supernatant (that contains unlabeled preexisting RNA) was placed at 4°C until 11 precipitation. Beads were washed and biotinylated RNA dissociated from streptavidin 12 coated beads by treatment with 100 mM 1,4-Dithiothreitol for 1 minute, followed by 5 13 minutes in RTL buffer. Beads were separated from the solution using DynaMag TM -2 14 Magnetic stand and the RNA recovered from the supernatant extracted using Qiagen

Identification of AGO2 bound regions in mESCs 9
Cutadapt [58] was used to remove sequence adapters from publicly available AGO2-10 CLIP sequencing reads from wild-type and Dicer mutant mESCs [30] . Trimmed reads 11 were mapped to the mouse genome (mm10) using bowtie [59] (bowtie -v 2 -m 10 --12 best -strata) as previously described [60]. Mapped reads from the same cell type were 13 merged AGO2 bound clusters identified using PARAlyzer v1. from different gene loci) were filtered out and the remaining reads summarised at a 31 gene level using an in-house script. Translational efficiency (TE) was calculated in R. 32 Briefly, raw genes ribosome footprints and total RNA counts were normalized using 33 the edgeR package to account for variable library depths (cpm function ; 1 [64]).Translational efficiency (TE) was calculated as the log2 ratio between normalized 2

RP counts and normalized TR counts. 3
Conserved short open reading frames within lncRNA transcripts were identified by 4 overlapping lncRNA loci with regions with positive phyloCSF scores, those that likely 5 represent conserved coding regions, in any of the three possible reading frames on 6 the same strand as the lncRNA transcript ([41]

RNA stability 4
Transcription was inhibited by adding Actinomycin D resuspended in Dimethyl 5 Sulfoxide at a final concentration of 10 µg/ml in supplemented mESC growth medium. 6 Stability of transcripts was inferred by comparing relative gene expression levels 7 (normalized to Actin-β) in cells incubated for 8 hours with Actinomycin-D and untreated 8

cells. 9 10
Candidate lncRNA and mRNA analysis 11 Enhanced Green Fluorescent Protein gene (see Supplementary Table S3) was 12 amplified from the pBS-U6-CMV-EGFP plasmid [65] with primers complementary to 13 EGFP and NheI restriction sites (see Supplementary Table S3)

DNA ligase according to manufacturer instructions. Plasmid was transformed into 16
DH5α subcloning efficiency bacterial cells and Sanger sequencing was used to 17 confirm correct orientation of EGFP insertion into plasmid (GFP). 18 19 lncRNA-c1 was amplified from mESC cDNA using sequence specific primers with 20 overhangs containing restriction sites for either XhoI or EcoRI (Supplementary 21 Table S3) and cloned directionally into XhoI-EcoRI digested pcDNA3.1(-) plasmid 22 to generate lncRNA-c1 construct downstream of T7 promoter. Ligation was 23 performed using T4 DNA ligase according to manufacturer instructions. GFP-24 lncRNA-c1 construct was generated adopting same cloning strategy but inserting 25 lncRNA-c1 into GFP construct. Sanger sequencing was used to confirm correct 26

sequence. 27
Cdkn1a was amplified from mESC cDNA using sequence specific primers with 28 overhangs containing restriction sites for either XhoI or EcoRI (Supplementary 29 Table S3) and cloned directionally into XhoI-EcoRI digested pcDNA3.1(-) plasmid 1 downstream of T7 promoter. Forward primers Cdkn1aDATG introduce a missense 2 mutation that deletes the 1 st position of the Cdkn1a start codon (Supplementary 3   Table S3). Ligation was performed as previously described and the correct 4 sequence of constructs was confirmed by Sanger sequencing. 5 2 MRE on GFP-lncRNA-c1 were mutated using, the Takara In-fusion HD cloning kit 6 according to manufacturer's instructions. The primers were designed using the 7 manufacturer online design tool (https://www.takarabio.com/learning-8 centers/cloning/in-fusion-cloning-tools) and are available in Supplementary Table  9 S3. The MREs were mutated sequentially using primer containing the mutation of 10 interest and AmpHiFi PCR Master Mix. PCR products were gel purified, ligated 11 using the In-fusion HD enzyme and transformed into Stellar competent bacterial 12 cells according to manufacturer' instruction. MRE mutation was confirmed through 13 Sanger sequencing. 14 1 MRE on GFP-lncRNA-c1 was mutated using the Phusion High-Fidelity 15 Polymerase. Briefly, primers containing a scrambled sequence of the seed region 16 within the MRE and wings complementary to the targeted sequence (Supplementary 17 For miRNA mimic and inhibitor transfections, mmu-miRNA294-3p inhibitors (30 mM), 5 mmu-miR294-3p, mmu-miR295-3p mimics (100 mM) and miRNA mimic negative 6 controls were transfected 24 hours after plasmid transfection using the RNAimax 7 transfection reagent according to manufacturer's instructions. RNA was extracted 24 8 hours after small RNA transfection and reverse transcription was performed according 9 to manufacturer's instructions as described above. 10

RNA immunoprecipitation 11
RNA immunoprecipitation was performed as previously described [66]. Briefly, 12 4.8x10 6 E14 WT cells were seeded into 10cm dishes 16 hours prior to harvest. At the 13 same time, 60µl of Protein A/G Plus-Agarose beads were incubated with 10µl of 14 Rabbit anti-AGO2 or 2.5µg of normal rabbit IgGs. Protein content in cell lysate was 15 split in half, adjusted to 1ml using IP Lysis buffer and supplemented with protease 16 inhibitors and RNase inhibitors. 50µl of diluted cell lysates were collected for Input 17 (5%). The remaining cell lysate was added to the A/B or IgG coupled beads and 18 incubated overnight at 4 o C, on a rotating wheel. After washing, 100 µl of RIP buffer + 19 1 µl of RNAse inhibitor was added to the beads and centrifuged. 20 µl and 80 µl of 20 supernatant were collected for protein and RNA analysis. Immunoprecipitated and 21 input RNA was extracted using TRIzol reagent, resuspended in DNAse reaction mix 22 (16µl ddH2O, 2µl 10x RQ1 DNase buffer, 2µl RQ1 DNase) and reverse transcribed 23 using the GoScript RT Kit and oligo d[T]18. RT-qPCR were performed as described 24 above. 25 10 µl of RIP supernatant and input samples were separated on 8% SDS-PAGE gels 26 and transferred to PVDF membranes. After incubation with 5% skim milk in 27 1xPBS/0.1% Tween-20, the membranes were washed and incubated with antibodies 28 against AGO2 (ARGONAUTE 2 Rabbit mAb) and Dicer (Rabbit anti-Dicer) at 4 o C. for 29 16h. Secondary antibodies (anti-Rabbit IgG-HRP) were incubated on membranes for 30 1h at RT at a dilution of 1:5000. Immunoblots were developed using the SuperSignal 1 West kit and detected using an imaging system. Membrane stripping was performed 2 by low pH method and AGO2 membrane was re-probed with antibodies against AGO2 3 (Argonaute 2 Mouse mAb). All membranes were stained using a coomassie blue 4 staining solution to ensure equal loading. 5 6 Data and Code Availability 7 RNA sequencing data was analyzed as described in Method Details; the data files are 8 available in the Gene Expression Omnibus accession number GEO: GSE143277. 9 Unprocessed Western blot images are available at Supplementary File 1 and 2. 10 11

Acknowledgements 12
We would like to thank the Genomics Technology Facility at the University of 13 Lausanne for help with RNA integrity analysis, library preparation and sequencing. in the 3'unstralated regions of AGO2 bound mRNAs (AGO2 cluster>0) whose 31 degradation rates were either significantly (n=711, red) or not significantly changed 32 (n=1127, black) between KO and WT cells, based on the 10 minutes pulse estimates.

33
(F) Distribution of the fold-change after 8 days of tamoxifen treatment in degradation 34 rate (estimated based on the 10 minutes pulse) of mRNAs (n=29900, red), cytosolic 35 (n=474, blue) and nuclear (n=2348, grey) lncRNAs, in KO relative to WT cells.

36
Statistics: *-p<0.05, **-p<0.01 and ***-p<0.001. c1ΔMRE (x-axis) in miRNA-depleted cells (KO) relative to WT mESC (y-axis). 10 Transcript expression was first normalized by the amount of Act-b and PolII and next 11 by the total amount of transfected vectors per cell estimated based on the levels of 12 relative Neomycin expression. Each point corresponds to the results of one 13 independent biological replicate. Statistics: *-p<0.05, **-p<0.01 and ***-p<0.001 two-14 tailed paired t-test. regions of mRNAs bound (AGO2 cluster>0) whose degradation rates were either 1 significantly (red) or not significantly changed (black) between KO and WT cells, based 2 on the 15 minutes pulse estimates. (G) Distribution of the fold-change after 8 days of 3 tamoxifen treatment in degradation rate (estimated based on the 15 minutes pulse) of 4 mRNAs (red), cytosolic (blue) and nuclear (grey) lncRNAs, in KO relative to WT cells.

5
(H) Distribution of the fold-change after 8 days of tamoxifen treatment in synthesis rate 6 of mRNAs (red), cytosolic (blue) and nuclear (grey) lncRNAs, in KO relative to WT 7 cells. Results for the 10-and 15-minutes pulse are presented separately. 8 9 Supplementary Figure S3-(A) Genome browser view of the region encompassing lncRNA-c1-MREΔ (x-axis) in wild-type mESC (WT). Transcript expression was first 1 normalized by the amount of Act-b and PolII and next by the total amount of 2 transfected vectors per cell estimated based on the levels of relative Neomycin 3 expression. Each point corresponds to the results of one independent biological 4 replicate. Statistics: *-p<0.05, **-p<0.01 and ** *-p<0.001 two-tailed paired t-test. 5 6 7 8

Figure S3
A.