DNA damage response inhibition at dysfunctional telomeres by modulation of telomeric DNA damage response RNAs

The DNA damage response (DDR) is a set of cellular events that follows the generation of DNA damage. Recently, site-specific small non-coding RNAs, also termed DNA damage response RNAs (DDRNAs), have been shown to play a role in DDR signalling and DNA repair. Dysfunctional telomeres activate DDR in ageing, cancer and an increasing number of identified pathological conditions. Here we show that, in mammals, telomere dysfunction induces the transcription of telomeric DDRNAs (tDDRNAs) and their longer precursors from both DNA strands. DDR activation and maintenance at telomeres depend on the biogenesis and functions of tDDRNAs. Their functional inhibition by sequence-specific antisense oligonucleotides allows the unprecedented telomere-specific DDR inactivation in cultured cells and in vivo in mouse tissues. In summary, these results demonstrate that tDDRNAs are induced at dysfunctional telomeres and are necessary for DDR activation and they validate the viability of locus-specific DDR inhibition by targeting DDRNAs.

T he DNA damage response (DDR) is a coordinated signalling network involving a large number of interacting proteins controlled by post-translational modifications and includes signalling, cell-cycle checkpoint enforcement and DNA repair 1,2 . Following DNA double-strand break (DSB) generation, the apical DDR kinase ataxia telangiectasia mutated (ATM) undergoes activation and phosphorylates the histone H2AX at serine 139 (named gH2AX); this event is necessary for the recruitment of DDR proteins to sites of DNA damage, including ATM phosphorylated at serine 1,981 (pATM) itself and p53-binding protein 1 (53BP1). Recently, a class of small non-coding RNAs (ncRNAs), termed DNA damage response RNAs (DDRNAs), has been shown to be generated on transcription of the damaged locus following DSB induction and then processed by the endoribonucleases DICER and DROSHA [3][4][5] . DDRNAs are dispensable for the direct recognition of DNA damage and gH2AX phosphorylation, but necessary for the secondary recruitment of DDR proteins to DSBs to form the so-called DDR foci 6 . A similar set of small ncRNAs, named damage-induced RNAs, has been shown to be involved in DNA repair by homologous recombination (HR) in plants and human cells 7,8 , and by non-homologous end-joining (NHEJ) in plants 9 .
Telomeres are the nucleoprotein complexes located at the tips of eukaryotic chromosomes, composed of repetitive DNA (TTAGGG in vertebrates), and coated by a set of proteins collectively known as the shelterin complex 10 . Dysfunctional telomeres resemble DSBs and they have been observed during ageing, cancer and a number of medical conditions [11][12][13][14][15][16][17] . Apart from telomeric repeat-containing RNA (TERRA), a non-coding UUAGGG-rich transcript starting from promoters located in the subtelomeric region 18,19 , in mammals no other transcripts at telomeres have been characterized so far.
Here we provide evidence that both strands of deprotected telomeres are transcribed to generate telomeric DDRNAs (tDDRNAs) and their precursors, whose inhibition leads to a reduction in the DDR activation at dysfunctional telomeres in living cells and in vivo.

Results
Telomere deprotection triggers both strands' transcription. To explore the potential generation of DDRNAs and their role at dysfunctional telomeres, we used Trf2 conditional knockout mouse embryonic fibroblasts (Trf2 F/ þ and Trf2 F/F mouse embryonic fibroblasts (MEFs)) carrying a Cre recombinase (Rosa26-CreERT2) inducible by 4-hydroxytamoxifen (4OHT) 20 . TRF2 is a component of the shelterin complex and prevents telomeric DNA from being recognized as a DSB 10 . Recombination by Cre promptly and robustly induced telomere deprotection and DDR activation at telomeres in the Trf2 F/F cell line only (Supplementary Fig. 1a; ref. 20). DDRNAs are small double-stranded RNAs (dsRNAs) transcribed at DSBs and carry the sequence of the damaged genomic locus 21 . To study tDDRNAs we employed miScript, a quantitative PCR with reverse transcription (RT-qPCR)-based method 22 developed to detect and quantify small RNAs, and designed primers to selectively amplify either strands of potential tDDRNAs generated at deprotected telomeres: teloC (C-rich strand of tDDRNA) and teloG (G-rich strand of tDDRNA; Supplementary  Fig. 1b). Using a gel-extracted fraction of RNAs shorter than 40 nucleotides-thus, minimizing the contribution of much longer telomeric transcripts, including TERRA 18,19,23 -we observed an induction of both teloC and teloG RNAs (Fig. 1a), thus with the potential to form in vivo double-stranded tDDRNA molecules. The signals detected were not due to genomic DNA contamination, since very low signal was detectable in the absence of reverse transcription ( Supplementary Fig. 1c). To extend these observations beyond mouse cells or TRF2 genetic ablation only, we used T19 cells, a human HT1080 fibrosarcoma cell line derivative, expressing a doxycycline-induced FLAGtagged dominant negative form of TRF2 (TRF2 DBDM), whose expression leads to telomere deprotection and DDR activation ( Supplementary Fig. 1d,e; ref. 24). We observed that also in this cell system, tDDRNA levels increased following telomere deprotection ( Supplementary Fig. 1f). This indicates that tDDRNA induction mechanisms are conserved among species.
To further characterize the length and sequence of tDDRNAs, we devised and employed an innovative method for target enrichment of RNA, based on in solution capture of lowabundance small RNA species followed by next-generation sequencing, developed in our laboratories (Nguyen et al., in preparation). By this approach, we observed that telomere deprotection induced the accumulation of small RNA species generated from the transcription of both telomere strands, including the expected DDRNA size range products ( Fig. 1b;  Supplementary Fig. 1g). Interestingly, 20-23 nucleotide RNAs displayed a base bias at both 5 0 -and 3 0 -ends significantly different from the telomeric locus ( Supplementary Fig. 2a), suggesting a regulated processing.
We next sought evidence for longer RNA species-we named them damage-induced long ncRNAs (dilncRNAs)-potential precursors of tDDRNAs. For this purpose, we performed a strand-specific fixed-length RT-qPCR 25 that allows the quantitative amplification of repetitive sequences, such as telomeric repeats, but not of RNA molecules shorter than 40 nucleotides such as DDRNAs ( Supplementary Fig. 2b). In cells with deprotected telomeres, both teloC and teloG long transcripts were 10-to 20-fold more abundant than in control cells (Fig. 1c), indicating that telomere deprotection strongly induces telomere transcription on both strands. Although PCR amplification of UUAGGG repeat-containing RNA may also detect the wellcharacterized telomeric transcript TERRA 18,19 , C-rich telomere repeat-containing RNA has not been characterized in mammals. To generate additional evidence of the telomere transcription induced on telomere deprotection by an independent approach, and to dissect the sub-cellular localization of telomeric dilncRNAs, we employed single-molecule fluorescence in situ hybridization (smFISH) 26 . To visualize telomeres in Trf2 F/F MEFs, we stably expressed a green fluorescent protein (GFP)tagged version of the telomeric protein TRF1, which co-localized with gH2AX foci on telomere deprotection ( Supplementary  Fig. 2c,d). By using strand-specific DNA probes, nuclear focal signals of RNA transcripts with telomeric sequence were detected. Strikingly, on telomere deprotection, we observed a marked increase in the number and the intensity of such nuclear signals for both G-and C-rich telomeric RNA, which often co-localized with GFP-TRF1 signals, and much less with centromeres used as a negative control (Fig. 1d,e; Supplementary Fig. 2e). In control cells expressing TRF2, a weaker signal was detected only by the probe against teloG, which likely detects also TERRA transcripts, while none was detected by probing for the complementary strand. The observed signals were specific for RNA, since they were lost when RNase A was added before hybridization ( Supplementary Fig. 2f). This result further strengthens the evidence of transcriptional induction of both DNA strands at deprotected telomeres. involved in DDRNA biogenesis on DSB generation 21 . We therefore investigated their role in the processing of telomeric transcripts, and consequently in DDR activation at dysfunctional telomeres, by individually knocking down each of them in Trf2 F/F MEFs ( Supplementary Fig. 2g). We observed that silencing of DICER or DROSHA fully abolished tDDRNA induction following telomere deprotection (Fig. 2a). Conversely, in telomere-deprotected cells, telomeric dilncRNAs increased significantly in Drosha knocked-down cells (Fig. 2b). This accumulation is consistent with dilncRNAs being DDRNA precursors, which are then processed by DROSHA-an increase on Dicer knockdown could not be appreciated likely because Dicer knockdown leads to the accumulation of RNA intermediates too short to be detected by qPCR in this experimental setting.
Next, we aimed to study the impact of the DDRNA biogenesis inhibition on DDR activation through Dicer or Drosha knockdown. In MEFs exposed to ionizing radiation (IR), as reported in Number of spots per cell 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29  other cell lines 21 , Dicer and Drosha knockdown impaired 53BP1 and pATM foci formation, while gH2AX foci remained unchanged ( Supplementary Fig. 2h,i). This was not due to altered 53BP1 or ATM protein levels ( Supplementary Fig. 3a,b), consistent with the reported unaltered levels of chromosomal integrity genes and telomere maintenance factors observed in conditional knockout mouse models for Dicer or Dgcr8, a Drosha cofactor 27 . We then tested whether DICER and DROSHA have a role in DDR activation at dysfunctional telomeres. As expected, DICER or DROSHA loss did not affect gH2AX foci at deprotected telomeres (Fig. 2c,d). When we studied the activation of the DDR apical kinase ATM, we observed that the fraction of cells containing pATM foci, as well as foci of proteins phosphorylated by PI3-like kinases, including ATM on their consensus target sequence (pS/TQ), were strongly reduced on Dicer and Drosha knockdown (Fig. 2c,d). 53BP1 foci formation was not detectably affected on telomere deprotection in Dicer and Drosha knockeddown cells (Fig. 2c,d); this is consistent with the notion that 53BP1 recruitment to DDR foci is controlled by redundant mechanisms 28 and is most dependent on DICER and DROSHA only at early time points (10 min) following DNA damage generation 21 , while the genetic deletion of TRF2 does not allow such early time points studies. Unbiased identification and quantification of DDR foci by an automatic imaging software 29 confirmed that in both DICER and DROSHA-depleted cells, the number, as well as the intensity, of pATM and pS/TQ foci per cell were reduced, whereas gH2AX and 53BP1 foci were unaffected ( Supplementary Fig. 3c). Furthermore, in human T19 cells, both Dicer and Drosha knockdown reduced the recruitment of pATM and ATM kinase targets pS/TQ, to deprotected telomeres, but not gH2AX foci formation ( Supplementary Fig. 3d,e). Therefore, in both mouse and human cells the stable recruitment of activated ATM and its phosphorylated targets to deprotected telomeres requires DICER and DROSHA. DICER-dependent small ncRNAs generated at sites of DNA damage, damage-induced RNAs, have been proposed to be necessary for efficient repair by HR in Arabidopsis thaliana and human cells 7,8 . While DICER seems dispensable for NHEJ DNA repair in a human cell-based reporter system 8 , Dicer-like 3 has been involved in NHEJ in plants 9 . More recently, error-free DSBs repair by NHEJ have been shown to involve RNA transcripts 30 Fig. 4a,b). These results indicate that DICER and DROSHA, and likely their DDRNAs products, are involved in NHEJ DNA repair at dysfunctional telomeres in mammalian cells.
DDR at deprotected telomeres is mediated by tDDRNAs. Thus far, the demonstrated effects of DICER and DROSHA on DDR activation at deprotected telomeres do not exclude the possibility that such effects may be indirect, meaning mediated by the role of DICER and DROSHA in microRNAs (miRNAs) biogenesis and gene expression. To address this issue, we used an experimental set-up, in which we previously demonstrated 21 that the impact on DDR activation of DICER-and DROSHA-generated DDRNAs can be studied independently from the canonical translational effects of miRNAs. We thus used a mild detergent to transiently permeabilize cell membranes of Trf2 F/F MEFs, in which DDR was triggered by IR or by Trf2 genetic ablation and treated them with RNase A, or bovine serum albumin (BSA) as control. Both IR-and telomere deprotection-induced 53BP1 foci were sensitive to RNase A (Fig. 3a,b; Supplementary Fig. 4c,d), despite unaltered 53BP1 protein levels ( Supplementary Fig. 4e,f), while gH2AX foci were not affected. Thus, also at deprotected telomeres RNA is necessary for DDR foci maintenance. To test if DDR foci can reform in an RNA-dependent manner, permeabilised RNase A-treated, and telomere-deprotected, MEFs were incubated with RNA extracted from MEFs with either normal (uninduced) or deprotected (induced) telomeres, or yeast transfer RNA (tRNA) as control. We observed that only RNA from induced MEFs allowed the robust reformation of 53BP1 foci in RNase A-treated cells (Fig. 3c). We then characterized the RNA species required for DDR foci reformation and observed that RNA extracted from induced cells knocked down for Dicer or Drosha was unable to restore 53BP1 foci (Fig. 3d). This indicates that the RNA species responsible for DDR foci maintenance at dysfunctional telomeres are DICER and DROSHA products. To further demonstrate that telomeric DICER products are indeed the RNA molecules necessary for DDR activation at telomeres and are sufficient to allow telomeric DDR foci reformation following RNase A treatment, we generated small double-stranded tDDRNAs by incubating a long double-stranded telomeric, or control, RNA substrate with purified recombinant DICER ( Supplementary  Fig. 4g). We then added these in vitro-generated telomeric and control DDRNAs to telomere-deprotected, RNase A-treated MEFs. We observed that only telomeric, but not control, DICER products could restore 53BP1 foci formation (Fig. 3e). Importantly, we can exclude that this effect was mediated by miRNA-like translational inhibition by DICER products, since messenger RNAs, the potential targets, were degraded by RNase A treatment and the effects were observed within minutes from RNA addition. To further prove the crucial dependency of DDR signalling at telomeres on tDDRNAs, we chemically synthetized 21-nucleotide long dsRNAs with a telomeric or a control sequence and tested their activity in RNase A-treated Trf2-deleted MEFs. We observed that the addition of synthetic tDDRNAs was sufficient to allow the reformation of robust DDR signalling events, as indicated by the reformation of 53BP1 foci, while DDRNAs with a control sequence had no effect (Fig. 3f). Consistent with a role of tDDRNAs in 53BP1 foci reformation at dysfunctional telomeres, tDDRNAs localized to dysfunctional telomeres, as measured by proximity ligation assay (PLA) between GFP-tagged TRF1 and Alexa 647-conjugated to tDDRNAs, or control DDRNAs that localized to telomeres much less ( Fig. 3g; Supplementary Fig. 4h). In summary, these results demonstrate that tDDRNAs have a crucial role in DDR activation at dysfunctional telomeres.
Antisense oligonucleotides allow DDR inhibition at telomeres. The results described so far consistently support the notion of a crucial role of tDDRNAs in DDR signalling at deprotected telomeres. On the grounds of the above conclusions, we next tested whether DDR at telomeres could be affected by the direct inhibition of tDDRNA functions. Inhibitory antisense oligonucleotides (ASOs) are widely used to block the functions of RNA molecules, including ncRNAs and small RNAs such as miRNAs 34 by steric hindrance. We therefore designed ASOs complementary to either tDDRNAs (anti-teloG and anti-teloC) or an ASO with an unrelated control sequence. We observed no obvious increase in gH2AX, 53BP1, pATM and pS/TQ foci in wild-type MEFs transfected with any of the ASOs used, ruling out the possibility that ASOs could compete with telomeres for the binding of telomeric proteins ( Supplementary Fig. 5a,b). ASOs were then transfected in MEFs before telomere deprotection, or exposure to IR as a source of non sequence-specific DNA damage, and DDR activation was monitored. ASO treatment did not have any impact on IR-induced DDR foci ( Supplementary Fig. 6a). Differently, both anti-teloG and anti-teloC, but not the control ASO, reduced both the number and the intensity of 53BP1 and pS/TQ foci at deprotected telomeres (Fig. 4a,b). Similar to Dicer and Drosha knockdown, and RNase A treatment, no effect on gH2AX foci was observed.
To test the direct role of tDDRNAs on chromosomal fusions following TRF2 depletion, we inhibited them by ASOs in cultured MEFs and monitored fusion events. We observed a significant reduction in the frequency of chromosomal fusions in cells treated with both anti-teloG and anti-teloC ASOs, but not with the control ASO ( Supplementary Fig. 6b).
These results demonstrate, for the first time, that DDR activation and NHEJ events at dysfunctional telomeres can be modulated in living cells by ASOs.
Telomeric DDR inhibition in vivo. ASOs have been successfully used to treat a number of different diseases; many of them are being tested in clinical trials and some have already been approved for use in the clinics 35,36 . To validate the efficacy of anti-teloG and anti-teloC ASOs in a living organism, we used an inducible Trf2 knockout mouse model in which, following tamoxifen administration, TRF2 expression is lost, leading to telomere deprotection and a very robust telomeric DDR activation (Fig. 5a,b). After tamoxifen administration, but before detectable DDR induction, mice received a systemic dose of anti-teloG or anti-teloC ASOs, or phosphatebuffered saline (PBS) as control, by intraperitoneal (i.p.) injection ( Supplementary Fig. 6c). Four days later, when telomeric DDR was expected to reach its maximal activation, organs were collected and tissue sections studied for DDR activation by immunofluorescence. We observed a marked reduction in foci intensity of both 53BP1 and the ATM target pKAP1 (ref. 37) in mice treated with both ASOs against tDDRNAs, compared with the control-treated mice, while gH2AX foci were unaffected. This striking effect could be observed in both organs studied: liver (Fig. 5a,b) and kidney (Fig. 5c,d). In summary, these unprecedented results demonstrate that sequence-specific targeting of tDDRNAs by ASOs allows telomeric DDR inhibition in both cultured cells and living mammals.

Discussion
Our results show that dysfunctional telomeres are actively transcribed to generate double-stranded telomeric dilncRNAs and DDRNAs. While G-rich transcripts (TERRA) have previously been widely studied, this is the first report on the generation, processing and function of C-rich telomeric transcripts in mammals to our knowledge. Very low-abundant signals were detected for C-rich telomeric RNAs also in ref. 18. The observation that Schizosaccharomyces pombe strains deleted for telomeric proteins Taz1 or Rap1 (yeast homologues of the mammalian shelterin components TRF1/2 and RAP1, respectively) also show an induction of C-rich transcripts 38,39 suggests that the observed induction is evolutionarily conserved. These results are consistent with RNA polymerase II recruitment at DSBs sites in fission yeast 31 , and with the reported increased RNA polymerase II occupancy at telomeres on TRF2 downregulation in human cells 40 . The mechanism by which the proteins involved in the DDR are recruited to dysfunctional telomeres in a tDDRNA-dependent manner remains to be further investigated, although the ability of 53BP1 to bind RNA through its Tudor domain 41 can be involved.
The evidence that only DDRNAs bearing a telomeric sequence are able to restore DDR activation at telomeres following RNase A treatment, and that only ASOs against tDDRNAs inhibit DDR at telomeres, but not at DSBs randomly generated at various genomic locations by IR, strongly suggests that tDDRNAs act in a locus-and sequence-specific manner.
Finally, we propose that the ability to detect tDDRNAs and to inhibit telomeric transcripts in vitro and in vivo by ASOs may have important implications in human physiology and pathology. tDDRNAs may be important biomarkers of DDR activation on telomeric damage, an event associated with physiological ageing and cancer [11][12][13][14][15][16] . In addition, many medical conditions, collectively known as telomere syndromes, and some progeric syndromes, are caused by accelerated telomere shortening and dysfunction with consequent DDR activation 17 . It is therefore conceivable that tDDRNA detection and modulation, for instance by ASO inhibition, can be exploited to address pathological conditions caused by telomere dysfunction.

Animals and treatments. Rosa26-CreERT mice (Jackson Laboratories) and
Trf2 conditional knockout mice 33 and mice carrying a p53 conditional allele (Jackson Laboratories) were crossed to generate Trf2/p53/Rosa26 mice. Mice were maintained in 129/c57Bl6 genetic background. All mice were bred and maintained under pathogen-free condition at the Scripps Research Institute, La Jolla, USA, and were handled according to Institutional Animal Care and Use Committee guidelines. To activate CreER, 8-10-week-old mice (both males and females) were injected i.p. with tamoxifen dissolved in sunflower oil or with vehicle at a final concentration of 75 mg kg À 1 . After 24 h, ASOs dissolved in PBS were administrated by i.p. at concentration of 15 mg kg À 1 . Mice were sacrificed after 5 days post tamoxifen injection. Tissues were collected and frozen in dry ice and embedded in optimal cutting temperature (OCT) tissue TEC (Sakura).
No power analysis was done to choose the size of the samples. No specific method of randomization was used to assign groups. Animals were assigned to experimental groups so as to minimize the influence of other variables such as age or sex on the outcome. No blinding approach was used for the experiments.
RNA isolation. Total RNA from cultured cells was extracted with RNeasy Mini Kit (Qiagen) for messenger RNA detection, or with mirVana miRNA Isolation kit (Life Technologies) for DDRNA and dilncRNA detection, according to the manufacturer's instructions. RNA extracted with mirVana miRNA Isolation kit using the Enrichment Procedure for Small RNAs was used as starting material for Target Enrichment experiments.
Real-time quantitative PCR. A measure of 1 mg of total cell RNA was reverse transcribed using SuperScript VILO cDNA Synthesis Kit. A volume corresponding to 10 ng of initial RNA was used for each qPCR reaction using GoTaq qPCR Master Mix (Promega) on a Roche LightCycler 480 sequence detection system. Each reaction was performed in triplicate. Human ribosomal protein large P0 (Rplp0) and mouse beta-2 microglobulin (B2M) were used as control transcripts for normalization. Primers sequences (5 0 -3 0 orientation) were: Analysis of small RNA sequencing data. Illumina adapters and linkers used in the targeted sequencing of small RNA were trimmed using Trim Galore! (http://www.bioinformatics.babraham.ac.uk/projects/trim_galore/), discarding reads that became shorter than 10 nucleotides of length after trimming. Reads were mapped to a telomeric contig made of 15 consecutive TTAGGG repeats and in parallel to the mouse genome (GRCm38/mm10) using Bowtie2 (ref. 42) with a very sensitive local option (-D 20 -R 3 -N 0 -L 20 -i S,1,0.50). Strand information was preserved to divide the aligned reads according to the original strand. Reads were parsed using SAMtools 43 and ad hoc bash and perl scripts. To avoid spurious reads contamination, telomeric reads with mismatches or soft clipped portions were removed with ad hoc R scripts.
RNA smFISH. Cells were fixed in 4% paraformaldehyde (PFA) for 10 min at room temperature (RT) and permeablized overnight at 4°C using 70% ethanol. Cells were rehydrated in a solution containing 20% formamide and 2 Â SSC for 5 min and then treated with 10 nM FISH probes for 12 h in 2 Â SSC containing 10% dextran sulfate, 2 mM vanadyl-ribonucleoside complex, 0.02% RNase-free BSA, 1 mg ml À 1 E. coli tRNA and 20% formamide at 37°C. After hybridization, the cells were washed twice for 30 min at 37°C using a wash buffer (20% formamide in 2 Â SSC). Cells were then mounted in solution containing 10 mM Tris/HCl, 2 Â SSC, 2 mM trolox, 5 mM protocatechiuc acid and 50 nM protocatechuate dehydrogenase. HILO imaging was done, as described 26 . FISH signals were detected using custom-written macros in ImageJ that were based on the analysis routine as described 44 . To calculate the relative intensity of FISH spots, a 21 Â 21 pixel area around the FISH signal locus was normalized with the intensity of a 21 Â 21 pixel area outside of the locus. Intensity of 21 Â 21 pixel area outside the cell was used for background subtraction. Red and green channels were corrected for signal bleed-through-samples that were labelled only with Cy5 or GFP were imaged using all excitation lines (488 and 640 nm) and signal arising from non-cognate excitation were used as a correction factor. All probes were tested for RNA specificity by visualizing the loss of FISH signal on RNase A treatment. For these control experiments cells were treated with RNase A (1 mg ml À 1 ) in DPBS at 37°C, before probe treatment. FISH probes were purchased as high-performance liquid chromatography purified DNA oligonucleotides from IDT. Probe sequences were as follows (5 0 -3 0 orientation): TeloG: CCTAACCCTAACCCTAACCCTAAC-Cy5 TeloC: GGTTAGGGTTAGGGTTAGGGTTAG-Cy5 Combined IF/RNA smFISH. MEFs Rosa26-CreERT2 Trf2 F/F eGFP-TRF1 were fixed in 4% PFA for 10 min and treated with a solution containing 1:1 methanol:acetic acid to irreversibly photobleach eGFP-TRF1. Cells were then washed three times in PBS for 5 min each and the standard immunofluorescence protocol was continued. In addition, all blocking, primary and secondary antibody incubations were performed in 1 Â PBBR (1% Ultrapure/RNase-free BSA þ 0.1 U ml À 1 rRNAsin, 1 mM EDTA in PBS). Importantly, the Human-anti-centromere antibody was purified by passing through G25 spin column (Millipore), which was pre-equilibrated in 1 Â PBS with 5% glycerol and 0.1 U ml À 1 rRNAsin. Cells were fixed again in in 4% PFA (in 1 Â PBS) for 10 min after immunofluorescence protocol to stably retain the bound primary and secondary antibodies and the standard RNA smFISH protocol was continued.
Transfection. Transfections were carried out with Lipofectamine RNAiMAX (Invitrogen) according to the manufacturer's instructions. Immunoblot. Cells were lysed in Laemmli sample buffer (2% SDS, 10% glycerol, 60 mM Tris-HCl pH 6.8). A measure of 50 mg of whole cell extracts were resolved by SDS-polyacrylamide gel electrophoresis. Proteins were transferred to nitrocellulose membrane, which was blocked in 5% milk in TBST (0.1% Tween in tris-buffered saline), and incubated with the primary antibody for 1 h at RT, and with a horseradish peroxidase-conjugated secondary antibody for 1 h at RT. Quantification of protein bands was done by ImageJ software, subtracting the background signal and normalizing for the housekeeper. Uncropped scans of the blots are provided in Supplementary Fig. 7.
Immunofluorescence for cultured cells. Cells were fixed with 4% PFA or 1:1 methanol/acetone solution. After incubation with blocking solution, cells were stained with primary antibody for 1 h at RT, washed and incubated with secondary antibodies for 40 min at RT. Nuclei were stained with 4,6-diamidino-2-phenylindole (DAPI; 1 mg ml À 1 ). Samples were mounted in mowiol or in imaging solution as in RNA smFISH.
Immunofluorescence for mouse tissues. A measure of 4-mm tissue sections were fixed for 10 min in 4% PFA, incubated in blocking solution (2% BSA, 0.1% Tween in PBS) for 1 hour at RT. Then sections were incubated for 1 h at RT with primary antibodies, washed in blocking solution and incubated for 1 h at RT with secondary antibody. Nuclei were stained with DAPI (1 mg ml À 1 ). Samples were mounted with glycerol solution.
Metaphase spreads. MEFs were incubated for 2 h with 0.2 mg ml À 1 colcemid (Gibco). The cells were collected by trypsinization, resuspended in 0.075 M KCl at 37°C for 30 min, and fixed overnight in methanol/acetic acid (3:1) at 4°C. The cells were dropped onto glass slides and the slides were dried overnight. The next day, the slides were rehydrated with PBS for 5 min. Slides were incubated consecutively with 75, 95, and 100% ethanol and allowed to air dry before FISH.
DNA FISH. Cultured cells after immunofluorescence were fixed with 4% PFA, 0.1% Triton X-100 for 10 min at RT, then incubated with 10 mM glycine for 30 min at RT. Interphase or metaphase cells were incubated with a telomeric PNA probe (0.5 mM, To remove salts and undigested templates, RNA Purification Column 1 and 2 (Genlantis) were used. RNA products were quantified and checked on a 3% agarose gel. As a control, the same procedure was followed with a 1 kb construct containing the neomycin DNA sequence. Equal amounts of DICER RNA products generated in this way were used in complementation experiments with MEFs following RNase A treatment.
Synthetic DDRNAs. 3 0 -labelled Alexa Fluor 647 RNA oligonucleotides were resuspended in 1 Â siRNA Buffer (Dharmacon). Equal amount of two oligonucleotides were mixed to generate double-stranded DDRNAs and treated as follows: 90°C for 1 min, 70°C for 2 min, 55°C for 2 min, 37°C for 5 min, RT for 5 min. Sequences were as follows (5 0 -3 0 orientation): Control pair UGUUAGCUGGAGUGAAAACUU GUUUUCACAAAGCUAACACA Telomeric pair AGGGUUAGGGUUAGGGUUAGG UAACCCUAACCCUAACCCUAA Statistical analysis. Results are shown as mean ± s.e.m. or s.d. or percen-tages±95% confidence interval as indicated. P value was calculated by the indicated statistical tests, using Prism software. Statistical significance of enrichment of small RNA species on telomere deprotection was calculated using the Fisher's exact test. The differences in the fraction of 20-23 nucleotides versus total small RNAs were calculated by performing a one-tailed binomial test, where the null hypothesis is the fraction of 20-23 nucleotide reads in the control sample and the alternative hypothesis is the count of 20-23 nucleotide reads versus total small RNAs in the sample with telomere deprotection. The deviation from a theoretically expected distribution in a UUAGGG or CCCUAA context was calculated by performing a one-tailed binomial test, where the null hypothesis is the expected fraction of U, A, G (or A, U, C) nucleotides in the wild type (that is, 2/6, 1/6 and 3/6, respectively), and the alternative hypothesis is the observed proportion of each nucleotide.
In figure legends, n indicates the number of independent experiments.
Data availability. The authors declare that all data supporting the findings of this study are available within the article and its Supplementary files or from the authors on a reasonable request. The sequencing data refer to the targeted sequencing of small RNA ( Fig. 1b; Supplementary Figs 1g  In Supplementary Fig. 3d, the two columns of DAPI images associated with pATM and pS/TQ images were inadvertently swapped. The correct version of this figure appears below.