PML induces compaction, TRF2 depletion and DNA damage signaling at telomeres and promotes their alternative lengthening

The gradual shortening of telomeres during replication eventually triggers growth arrest and senescence and thus provides an important tumor suppressor mechanism (d'Adda di Fagagna et al., 2003; Harley et al., 1990). Cancer cells overcome this proliferation limit by activating a telomere maintenance mechanism. In most cases telomerase is re-activated, which can extend the telomere repeat sequence TTAGGG (Shay and Bacchetti, 1997). However, 10–15% of tumors employ an alternative lengthening of telomeres (ALT) mechanism to elongate their chromosomal ends by DNA recombination and repair processes in the absence of telomerase (Bryan et al., 1997). ALT tumors are typically characterized by a large heterogeneity in telomere length within one cell (Bryan et al., 1995), the occurrence of extrachromosomal telomeric repeats (ECTRs) (Wang et al., 2004), mutations of the chromatin remodeler ATRX (Heaphy et al., 2011), genome instability (Lovejoy et al., 2012), increased telomeric recombination (Londoño-Vallejo et al., 2004) and the presence of ALT-associated promyelocytic leukemia (PML) nuclear bodies (APBs) (Chung et al., 2012; Yeager et al., 1999). APBs are defined as complexes of PML nuclear bodies (PML-NBs) with telomeric DNA in telomerase-negative cells (Yeager et al., 1999), and their ectopic assembly in ALT-positive cells induces telomere lengthening by promoting repair-associated DNA synthesis (Chung et al., 2011). A number of proteins involved in ALT have been identified, such as the telomeric shelterin complex (Jiang et al., 2007), the small ubiquitin-like modifier (SUMO) E3 ligase MMS21 (also known as NSMCE2) (Potts and Yu, 2007), several DNA repair proteins (Nabetani and Ishikawa, 2011), as well as heterochromatin protein 1 (HP1) family proteins (Jiang et al., 2009). However, the molecular details of the ALT pathway have remained elusive.

Here, we applied a three-dimensional (3D) confocal laser scanning microscopy (CLSM) screening platform and quantitative image analysis to evaluate changes in the nuclear organization of APBs, PML-NBs and telomeres at high precision based on the analysis of more than 20 million images. With this approach, we were able to characterize features of single telomeres in their native cellular context and compare the effect of small interfering RNA (siRNA)-mediated knockdown of ∼100 genes by analyzing a comprehensive set of image-based readouts. Our results reveal that depletion of APBs by long-term PML knockdown leads to telomere shortening and a reduction of ECTRs. In addition, we found that PML induced clustering and compaction of colocalizing telomere repeats and, simultaneously, reduced binding of the telomeric repeat binding factor 2 (TRF2, also known as TERF2). These changes in telomere organization correlated with the activation of the ataxia telangiectasia mutated (ATM) kinase in APBs. Based on these findings, we propose a model for APB-mediated telomere lengthening in ALT-positive cells and tumors.

PML is the central structural component for forming PML-NBs and APBs. In the absence of PML, other PML-NB components, such as SP100 and SUMO, do not assemble into a nuclear subcompartment (Ishov et al., 1999; Tavalai et al., 2006; Zhong et al., 2000). Accordingly, we investigated the role of PML protein in ALT by using an ALT-positive human U2OS osteosarcoma cell line with an inducible stable knockdown of PML that targets a sequence common to the seven PML isoforms. Immunofluorescence analysis using a pan PML antibody that detects all isoforms was conducted and evaluated with our previously developed quantitative automated 3D confocal image acquisition and analysis platform (Osterwald et al., 2012; Wörz et al., 2010) (supplementary material Fig. S1). The results showed that the number of PML-NBs was reduced by 99.3±0.1% (mean±s.e.m.) after 72 h of PML knockdown (P<0.001, Fig. 1A). The knockdown completely suppressed colocalizations between PML and telomeres and thus APB formation (–3.5±0.3 APBs per cell and –99.7±0.3%, respectively, P<0.001, Table 1). The disappearance of APBs was accompanied by a reduction in the amount of C-circles, which are ALT-specific partially single-stranded telomeric (CCCTAA)_n DNA circles (–87.5±4.1%, P<0.001, Fig. 1C). At the same time, the number of detectable telomere repeat foci per cell was significantly increased (+6.5±2.6, P<0.001, Table 1). The reduced fluorescence intensities of the Cy3-labeled telomere repeats after 72 h revealed that high-intensity telomere repeat signals disintegrated into several low-intensity telomere repeat foci (median, –12.8±2.0%, P<0.001, Fig. 1D). Thus, on average, the number of detectable telomere repeat foci increased upon PML knockdown by one to two for every APB that disappeared, indicating telomere clustering in APBs.

To assess whether PML is needed for telomere elongation in ALT cells, we performed a long-term PML knockdown in U2OS cells for 30 days, which corresponds to approximately 30 population doublings. The knockdown led to a significant decrease of the telomere repeat signal intensity (median, –24.9±1.7%, P<0.001, Fig. 1B,D) as detected by interphase quantitative fluorescence in situ hybridization (Q-FISH), which was more pronounced than the short-term effect observed after 72 h of knockdown. This reduction of telomere content upon long-term PML knockdown was confirmed by telomere-repeat quantitative PCR (supplementary material Fig. S2A). In addition, terminal restriction fragment (TRF) analysis after 2, 4 and 6 weeks of PML knockdown revealed that PML knockdown induced telomere shortening (supplementary material Fig. S2B). Next, we performed Q-FISH on metaphase chromosomes of uninduced and induced PML knockdown cells (Fig. 1E; Table 2). This method is well established to detect and quantify ECTRs and has been used in a number of previous studies (Episkopou et al., 2014; Hande et al., 2001; Kamranvar et al., 2013; Kamranvar and Masucci, 2011; Tokutake et al., 1998). Note that we refer here to those ECTRs that are detected by a peptide nucleic acid (PNA) FISH probe against the G-rich telomere repeat sequence. This group of ECTRs is distinct from the single-stranded C-rich C-circles measured by rolling circle amplification according to the method of Henson et al. (Henson et al., 2009), which would not give rise to a signal in our telomere-repeat FISH assay. The intensity of telomere repeats associated with chromosomes was significantly reduced after 30 days of PML knockdown (median, –18.6±9.6%, P<0.001, Table 2), and considerably more telomere-free ends were detected (52.5±26.5%, P<0.05, Table 2). The number of detectable ECTRs – defined as telomere repeat signals per metaphase spread that were not associated with chromosomes – was reduced by 59.8±10.2% (P<0.001, Table 2). In general, ECTRs had a lower median repeat intensity than the telomeres. They accounted for only ∼7% of the total telomere repeat intensity per metaphase spread in uninduced control cells, thus representing a small fraction of the total telomere repeat signal (Table 2). This finding agrees well with a recent study that also quantified the contribution of ECTRs to total telomere repeat content in ALT cells by performing both Q-FISH and extraction of extrachromosomal DNA and subsequent Southern blotting (Episkopou et al., 2014). Accordingly, we conclude that the telomere repeat signal measured in interphase FISH experiments (Table 2) mainly originates from telomeres and only a small fraction represents ECTRs.

In summary, interphase and metaphase Q-FISH, TRF analysis and telomere quantitative PCR consistently reveal that telomere shortening is induced upon PML knockdown that is accompanied by a reduction of ECTRs, including both C-circles and G-rich ECTRs.

Having shown that depletion of APBs by PML knockdown led to telomere shortening, we set out to identify factors that disrupt PML assembly into APBs and thus could affect the ALT pathway. We conducted an RNA interference (RNAi) screen by quantitative automated 3D confocal imaging and subsequent analysis of telomere, PML-NB and APB features, as described in further detail previously (Osterwald et al., 2012; Wörz et al., 2010). Briefly, ∼100 candidate proteins were knocked down by two independent small interfering RNAs (siRNAs) (supplementary material Table S1). Then the number, volume, intensity and density (defined as intensity per volume) of telomere repeats and PML-NBs, and their colocalization, representing APBs, were determined from the automated analysis of more than 20 million images (supplementary material Fig. S1). In this manner, we were able to reliably quantify changes in APB formation and telomere organization at the level of single telomeres with high precision in order to dissect the function of APBs.

From our RNAi screen, we identified 29 proteins involved in APB formation (supplementary material Table S2). Only those proteins that showed a significant change of more than 10% in the number of APBs (P<0.05) for two different siRNAs in at least three independent experiments were selected as hits. Other ‘non-hit’ proteins that did not meet these relatively strict requirements were classified into two groups (supplementary material Table S2): (1) targets where both siRNAs consistently did not show a significant effect on the number of APBs, considered as proteins that do not have an effect on APB formation; and (2) targets where only one out of two siRNAs showed a significant effect, representing candidates that could have an effect on APB formation. However, these were not further investigated here. The knockdown efficiency of selected hits as well as non-hits was analyzed by quantitative real-time PCR (supplementary material Fig. S3A) unless already previously validated (supplementary material Table S1). Based on the associated biological processes according to gene ontology (GO) annotation, the APB effector proteins were grouped into proteins involved in telomere organization, protein sumoylation, DNA repair and chromatin organization (Table 3).

As inferred from our previous work, the number of APBs displayed little dependence on cell cycle state in normally proliferating U2OS cells (Osterwald et al., 2012), although the number of PML-NBs was higher during S-phase in this cell line (Dellaire et al., 2006). For other cell lines, a higher number of APBs has been reported after inducing a cell cycle arrest in G2/M phase for human ovarian surface epithelium (HOSE) cells (Grobelny et al., 2000) or in G0/G1 phase for IIICF/c and GM847 cell lines (Jiang et al., 2007). Accordingly, we addressed the question of whether the siRNA knockdowns conducted here were associated with significant changes in the proportion of cells in each phase of the cell cycle. The integrated background-corrected DAPI intensities per cell nucleus were computed for a given sample from the confocal image stacks and used to obtain the relative cellular DNA content as described previously (Tóth et al., 2004; Osterwald et al., 2012). From this quantification the cell cycle distribution was determined by applying an identical predefined gating of the DNA content histograms for G1, S and G2/M phase and computing the corresponding percentage of cells within each group (supplementary material Fig. S3B). For the majority of knockdown experiments no significant change in the proportion of cells in each phase of the cell cycle was observed under our experimental conditions (supplementary material Table S2). For six targets (BLM, CDKN1A, FEN1, LSD1, MORC3 and UBC9) a significant change, in the range of 10 percentage points for cells in G1 or G2/M, but not in S phase, was observed for one of the siRNAs in comparison to control siRNA. However, the direction of the measured change in the number of APBs for siRNAs targeting FEN1, UBC9 and CDKN1A did not correspond to that observed previously in cell cycle arrest experiments with IIICF/c, GM847 or HOSE cells (Grobelny et al., 2000; Jiang et al., 2007). Thus, we conclude that in general the changes in the number of PML-NBs and APBs measured here for the U2OS cell line upon siRNA-mediated protein knockdown were not due to changes in the cell cycle distribution.

Next, we evaluated the link between telomeric chromatin organization and APB formation. Interestingly, knockdown of several factors involved in heterochromatin formation, for example, the histone methyltransferase SUV4-20H2 (Benetti et al., 2007; Schotta et al., 2004), heterochromatin protein 1γ (HP1γ, also known as CBX3) (Jiang et al., 2011; Jiang et al., 2009; Verschure et al., 2005) or the histone demethylase LSD1 (Shi et al., 2004), reduced the number of APBs. In contrast, the number of APBs increased upon RNAi-mediated knockdown of high-mobility group nucleosome binding domain 5 (HMGN5), which counteracts the chromatin-condensing activity of linker histones (Rochman et al., 2009). Thus, the targeting of various chromatin modifiers had a significant effect on APB formation.

This prompted us to further investigate the role of the telomeric chromatin state in the ALT pathway. We evaluated differences in telomere compaction as reflected by the telomere repeat density. This parameter was derived from telomere FISH images by dividing the intensity of a Cy3-labeled focal telomere repeat by its volume. To evaluate whether changes in telomere repeat compaction affected TRF2 binding to telomeres, we determined the ratio of TRF2 density to telomere repeat density from colocalizing TRF2 immunofluorescence and telomere FISH signals. The measured TRF2 to telomere repeat ratio was then corrected for the slightly reduced accessibility of the TRF2 antibody to telomere repeats of higher density as described in supplementary material Fig. S4. In this manner, we were able to compare TRF2 binding to telomere repeats at specific telomeres within one cell for different treatments. Next, we induced chromatin decondensation by treatment with the histone deacetylase inhibitor SAHA (Bradner et al., 2010; Choudhary et al., 2009; Tóth et al., 2004). Treatment with SAHA significantly reduced the telomere repeat density (P<0.001, Fig. 2A). The observed reduction in telomere repeat density after SAHA treatment was accompanied by a small, but statistically significant, increase in TRF2 binding per telomere repeat (P<0.01, Fig. 2B). Furthermore, SAHA strongly decreased the number of APBs per cell (mean±s.e.m., –49.4±6.9%, P<0.001, Fig. 2C) and C-circle levels (–45.3±10.3%, P<0.001, Fig. 2D).

We next employed a previously introduced technique to induce the de novo formation of ectopic APBs by recruiting GFP-tagged proteins to three telomeres in U2OS cells with stably integrated lac operator (lacO) arrays (Chung et al., 2011; Jegou et al., 2009). As controls, GFP alone was recruited (Fig. 2E) and a cell line with pericentric lacO array integration sites was used. HMGN5, as a factor that decondenses chromatin, and HP1γ, as a protein that promotes heterochromatin formation, were recruited (Fig. 2F,G). The capability of the two proteins to promote APB formation was monitored by the enrichment of endogenous PML protein at the telomeric lacO arrays. Recruitment of HMGN5 resulted in strong chromatin decondensation at the telomeric lacO arrays, as assessed by the formation of extended structures with irregular shape, which has previously been reported for non-telomeric lacO arrays (Rochman et al., 2009). No APBs could be detected at the telomeres of these cells, whereas 24±5% of lacO sequences were associated with APBs in the control cells (P<0.001, Fig. 2F). In contrast, recruiting GFP–HP1γ to the telomeric lacO arrays induced the subsequent enrichment of endogenous PML protein at the telomeres in a highly efficient manner, yielding 76±11% colocalization (P<0.001, Fig. 2G). PML enrichment upon HP1γ recruitment could be due to SUMO-mediated interactions as discussed previously (Lang et al., 2010) or could occur through SP100 (Seeler et al., 1998), a known interaction partner of PML. Notably, PML enrichment was accompanied by the induction of repair-associated DNA synthesis, as concluded from the increased levels of the phosphorylated histone variant H2A.X (γH2A.X) and of incorporated 5-bromo-2'-deoxyuridine (BrdU) (Fig. 2H,I). This effect was specific for telomeric lacO arrays. Recruitment of HP1γ to pericentric lacO arrays had no significant effect on the level of γH2A.X (Fig. 2J). Taken together, decondensation of telomeric chromatin inhibited APB formation, whereas a compacted chromatin state was found to be compatible with both APB formation as well as repair-associated DNA synthesis at telomeres.

To investigate differences in the level of compaction at single telomeres in unperturbed ALT cells, we analyzed the telomere repeat density in APBs as compared to telomere repeat foci that were not located in APBs. The median telomere repeat density in APBs was 2.6±0.1-fold higher as compared to telomeres outside APBs (P<0.001, Fig. 3A). To distinguish whether APBs induce compaction of telomeric chromatin or whether they assemble at pre-existing highly dense telomeres, the effect of short-term PML knockdown on the telomere repeat density was evaluated (Table 1). The median telomere repeat density was significantly reduced by 9.9±3.4% (mean±s.e.m.) after PML knockdown. This indicates that the increased compaction of telomere repeats was induced by APBs and was maintained only as long as PML was present.

Next, we evaluated whether the APB-mediated increase of the telomere repeat density influenced TRF2 binding to these telomeres. The ratio of the TRF2 density to telomere repeat density was strongly decreased in APBs as compared to telomeres that were not located in APBs (median, –35.2±4.9%, P<0.001, Fig. 3B). Thus, telomeres in APBs had less TRF2 bound per telomere repeat. Interestingly, telomere repeat density was inversely correlated with TRF2 binding: the least-dense telomeres had 2.6±0.7-fold more TRF2 bound per telomere repeat as compared to the densest telomere (P<0.001, Fig. 3C). This value includes the above-mentioned correction for differences in antibody accessibility (supplementary material Fig. S4G). To confirm that PML was required for the reduced binding of TRF2 at telomeres in APBs, we measured TRF2 binding to telomere repeats before and after PML knockdown. Notably, PML knockdown increased the amount of TRF2 bound per telomere repeat by ∼40% (Table 1). No change in the mean integrated TRF2 immunofluorescence signal per cell was detected, indicating that TRF2 levels per cell remained unaffected. Thus, we conclude that APBs induce a compaction of telomeric chromatin that correlates with reduced binding of TRF2 per telomere repeat.

Next, we investigated whether knockdown of proteins involved in post-translational modifications of shelterin proteins affected the binding of TRF2 to telomeres in APBs. The ratio of TRF2 signal to the telomere repeat signal (i.e. the coverage of telomere repeats with TRF2) was affected inside APBs upon knockdown of two different post-translational modifiers of TRF2, which have been reported to be enriched in APBs (Dantzer et al., 2004; Potts and Yu, 2007). First, knockdown of the SUMO E3 ligase MMS21, which sumoylates the shelterin components TRF1, TRF2, TIN2 and RAP1 (Potts and Yu, 2007), increased TRF2 binding to telomeres in APBs by 18.1±1.9% (mean±s.e.m., P<0.001). The opposite effect on the ratio of the TRF2 signal to the telomere repeat signal was observed for the knockdown of the SUMO protease SENP6, whereas other SENPs had no effect or decreased TRF2 binding to telomere repeats. Second, knockdown of poly(ADP-ribose) polymerase 2 (PARP-2) increased the ratio of the TRF2 signal to the telomere repeat signal in APBs by 9.3±0.3% (P<0.001) without affecting the telomere repeat density. Thus, post-translational modifiers of TRF2 that are known to be present in APBs can affect the amount of TRF2 that is bound per telomere repeat in APBs.

TRF2 is the main repressor of DNA damage response (DDR) at telomeres because it inhibits the autophosphorylation of the ATM kinase and its concomitant dissociation into monomers, the presumed active form of the kinase (Bakkenist and Kastan, 2003; Takai et al., 2003). As TRF2 binding to telomeres was strongly reduced in APBs (Fig. 3B), we addressed the question of whether this leads to the activation of ATM by autophosphorylation. Consistent with a previous report (Stagno D'Alcontres et al., 2007), the phosphorylated form of ATM (p-ATM) colocalized with APBs in U2OS cells (Fig. 4A, row 1). A quantitative analysis revealed that p-ATM was significantly enriched at telomeres associated with PML (Fig. 4B). Although only ∼2% of all telomeres colocalized with p-ATM, this fraction was significantly increased among telomeres in APBs. Approximately 15% of all APBs colocalized with p-ATM, indicating that the DDR was only activated in a specific subset of APBs (Fig. 4B). The median telomere repeat density in APBs with p-ATM was ∼2.4 times higher than the density in APBs without p-ATM and even ∼5.7 times higher than at telomeres that were not located in APBs (Fig. 4C). As shown in Fig. 3C, telomeres with the highest telomere repeat density had the lowest levels of bound TRF2. Thus, we conclude that ATM was only activated at telomeres in APBs with the highest telomere repeat densities and the lowest TRF2 levels.

To distinguish whether APBs enrich phosphorylated ATM at telomeres or whether APBs form at telomeres where the DDR is already activated, we analyzed the p-ATM distribution after 1 week of PML knockdown. The knockdown of PML increased the total number of p-ATM foci (+102±7%, mean±s.e.m., P<0.001, Fig. 4A,D), which is likely to reflect the previously reported general role of PML in DNA repair (Bernardi and Pandolfi, 2007). However, whereas in control cells about a quarter of all p-ATM foci were found at telomeres, the number of p-ATM foci at telomeres was strongly reduced after PML knockdown (–90±8%, mean±s.e.m., P<0.001, Fig. 4D). This indicates that ATM becomes activated at telomeres after APB formation rather than inducing the formation of APBs at telomeres where the DDR was already initiated. In support of this conclusion, ATM knockdown had no significant effect on the number of APBs (supplementary material Table S2). As reported previously, ATM interacts with the MRN ‘damage sensor’ complex, which leads to the recruitment of other repair proteins like MDC1 and 53BP1 (Derheimer and Kastan, 2010). Although these proteins are known to colocalize with APBs (Jiang et al., 2007), their knockdown had no effect on the number of APBs in our screen (supplementary material Table S2), implying that a functional DDR was not necessary for APB formation.

Next, we evaluated whether APB-induced ATM phosphorylation was necessary for telomere elongation by inhibiting ATM for 4 weeks with KU-55933 (Hickson et al., 2004). This treatment reduced the number and density of p-ATM foci by 45.2±2.1% and 48.2±3.6%, respectively (P<0.001). Furthermore, the amount of C-circles was reduced after treatment with the ATM inhibitor (–33.6±10.8%, P<0.05, Fig. 4E). Notably, the median fluorescence intensity of the Cy3-labeled telomere repeats (–12.9±2.7%, P<0.001, Fig. 4F) decreased without significantly affecting the number of APBs per cell (+2.8±3.5%, P>0.1). Thus, ATM inhibition correlated with a loss of telomere repeats, but did not affect APB formation. This indicates that ATM activation in APBs promotes subsequent DNA-repair-associated telomere elongation.

In the present study, we investigated the link between APB formation, TRF2 binding to telomeres and telomere lengthening in the ALT-positive U2OS cell line. Based on our findings, we propose a model for APB-mediated telomere lengthening that involves the following main steps and integrates findings from previous studies (Fig. 5). First, formation of a PML subcompartment at telomeres induces telomeric chromatin compaction and clustering of telomeres, and possibly also ECTRs, as proposed previously (Cho et al., 2014; Draskovic et al., 2009). Second, as a result of APB formation, TRF2 becomes partly depleted at associated telomeres. This process could involve post-translational modifications of TRF2, such as sumoylation by MMS21 or poly-ADP ribosylation by PARP-2, in line with previous studies (Dantzer et al., 2004; Potts and Yu, 2007). Third, the reduced TRF2 density triggers autophosphorylation of ATM in APBs and DDR according to the previously identified role of TRF2 as an inhibitor of ATM (Denchi and de Lange, 2007; Karlseder et al., 2004). Fourth, telomeres are elongated by repair-associated DNA synthesis and recombination events that are promoted by telomere clustering in APBs. Finally, as APBs disassemble, repair and recombination factors dislocate, telomeres are released and telomere density decreases again. This process leads to a re-enrichment of TRF2 that protects the extended telomeres from chromosomal fusions by non-homologous end joining (NHEJ).

Several lines of evidence support this model. An important role of APBs was established from the quantitative evaluation of the effect of PML knockdown in the ALT-positive U2OS cell line. Short-term PML knockdown for 72 h led to an almost complete loss of APBs, whereas the number of detectable telomere repeat foci increased by one to two for every APB that disappeared. This suggests clustering of two or three telomere repeat foci in one APB. Our Q-FISH interphase analysis did not reveal whether the additional telomere repeat foci observed after PML knockdown were telomeres or ECTRs. However, Q-FISH with a C-rich PNA probe on metaphase spreads showed that ECTRs accounted only for a relatively small fraction of the total telomere repeat intensity per cell. Furthermore, the number of detectable ECTRs was strongly reduced after PML knockdown. Thus, we conclude that the additional telomere repeat foci that appear after 72 h of PML knockdown arise mostly from telomeres. However, it is possible that ECTRs also contribute to the telomere repeat clusters inside APBs. Consistent with the view that APBs promote telomere clustering, it has been reported that telomeres attach to the surface of artificially enlarged APBs (Draskovic et al., 2009) and that damaged telomeres preferentially cluster with telomeres that are associated with PML in ALT-positive cells (Cho et al., 2014). Notably, long-term PML knockdown induced telomere shortening and significantly increased the number of chromosomal ends where a telomere repeat signal was absent. This demonstrates that PML is crucial for telomere elongation in ALT cells and confirms previous conclusions (Chung et al., 2011; Jiang et al., 2005). Although inhibition of the ALT mechanism by other means has been employed previously (Jiang et al., 2005; Potts and Yu, 2007), our study is the first to reveal the crucial contribution of PML by showing a telomere shortening upon its knockdown.

Having established the importance of PML for the ALT mechanism, we investigated the formation of APBs and their function in the ALT pathway with an automated quantitative 3D image acquisition and analysis approach in conjunction with RNAi-mediated knockdown. The quantification of individual telomeres and APBs from a total of more than 20 million images allowed us to identify 29 factors involved in APB formation and to elucidate the subsequent effects on telomere organization with unprecedented precision. The mechanism by which these factors operate is likely to involve direct effects that promote telomeric APB assembly as well as indirect effects related to DNA damage and its repair. Cell cycle effects appear to be less relevant in this context given that only very few tested siRNAs had significant effects on the cell cycle distribution. Note that proteins where the two targeting siRNAs showed inconsistent effects were not considered as hits in our screen (supplementary material Table S2). However, these proteins might nevertheless be involved in APB formation and the ALT mechanism as exemplified by the ataxia-telangiectasia- and RAD3-related (ATR) protein for which only one out of two siRNAs significantly reduced the number of APBs in our study. Indeed, a recent paper has shown that knockdown or inhibition of ATR specifically inhibits the ALT pathway and also reduces the number of APB-positive U2OS cells (Flynn et al., 2015). Knockdown of other DNA repair proteins, like the MRN complex components Rad50 and NBS1, as well as MDC1, 53BP1, BRCA1 and RAD51, did not affect the number of APBs for both siRNAs used, indicating that functional DDR and DNA repair pathways are not essential for APB formation (supplementary material Table S2). For 53BP1 and RAD51 knockdown, this is consistent with previous reports (Jiang et al., 2007; Potts and Yu, 2007). With respect to knockdown of RAD50 and NBS1, there is a disagreement with a previous study that reported a reduction in APB-positive IIICF/c cells upon knockdown of MRN components (Jiang et al., 2007). One reason could be that the abovementioned work used methionine-restriction-induced cell cycle arrest to artificially enrich the number of APBs. This treatment per se could have an impact on either APB formation or ALT. Accordingly, the effect of protein knockdown might be different from what is observed under the conditions used here for U2OS cells. A role for DNA repair proteins downstream of APB formation is also supported by our previous findings (Chung et al., 2011). Some repair proteins were inefficient in inducing the de novo formation of APBs, but instead were recruited to pre-assembled APBs. Note that the above results do not exclude the possibility that DNA damage also promotes the assembly of PML at telomeres in ALT-negative cells as reported previously (Hsu et al., 2012; Slatter et al., 2012).

The role of the telomeric chromatin state with respect to APB formation and telomere elongation in ALT cells is controversial. In telomerase-positive mice, it has been reported that knockout of several chromatin modifiers involved in heterochromatin formation results in APB formation and increased recombination at telomeres (Benetti et al., 2007; García-Cao et al., 2004; Gonzalo et al., 2006). To what extent these finding apply to human cells is unclear, given the differences in telomere biology between humans and mice (Calado and Dumitriu, 2013). Furthermore, a number of findings demonstrate that induction of a condensed heterochromatic state can even promote DNA repair and/or homologous recombination (Ayrapetov et al., 2014; Geuting et al., 2013). A recent study of DDR signaling in U2OS cells is particularly informative on this issue (Burgess et al., 2014). It shows that chromatin compaction is an integral part of DDR signaling and follows a transient chromatin expansion step.

We found here that APB assembly in U2OS cells was inhibited by an ‘open’ telomeric chromatin state, as the knockdown of several repressive chromatin modifiers, as well as chromatin decondensation initiated by HDAC inhibition or HMGN5 recruitment, resulted in a significant reduction in the number of APBs (Fig. 2C,F). Previous work in ALT-positive IIICF/c cells has shown that HP1α (also known as CBX5) and HP1γ are needed for APB formation under methionine restriction and the authors hypothesized that HP1-mediated chromatin compaction is required for APB formation (Jiang et al., 2009). It was concluded that compacted telomeric DNA inside APBs would counteract telomere–telomere recombination. Here, we show that recruitment of HP1γ to telomeres is compatible with PML-induced DNA repair synthesis (Fig. 2H,I). This is in line with studies demonstrating the importance of HP1 family proteins for DNA repair and recombination, as discussed in several reviews (Cann and Dellaire, 2011; Dinant and Luijsterburg, 2009; Soria et al., 2012). Recently, it has been reported that chromatin compaction is globally reduced at ALT telomeres in comparison to telomeres in telomerase-positive cells (Episkopou et al., 2014). Our work focused on analyzing the compaction of single telomeres within an ALT cell line and has revealed differences in telomere repeat densities in relation to their association with PML (Fig. 3). In particular, we found that telomere repeats in APBs were more compact and bound less TRF2 than telomere repeats outside of APBs (Fig. 3B). Interestingly, the high telomere repeat densities observed in APBs correlated with the activation of a DNA damage response through ATM phosphorylation.

Previous reports have already speculated that partial telomere deprotection might be important for the repair-based ALT mechanism (Cesare et al., 2009; Cesare and Reddel, 2008; Nabetani et al., 2004). In particular, a lack of TRF2 at ALT telomeres has been proposed to be the cause of this deprotection, because the ratio of total TRF2 levels to the amount of telomeric DNA is significantly lower in ALT-positive cell lines compared to telomerase-positive cell lines (Cesare et al., 2009). Here, we specifically compared the ratio of TRF2 density to telomere repeat density as derived from colocalizing TRF2 immunofluorescence and telomere FISH signals at single telomeres in the ALT-positive U2OS cell line. This approach has allowed us to reveal differences in TRF2 binding to telomeres with or without APBs. Based on this comparison and the fact that PML knockdown led to reduced telomere repeat density and increased binding of TRF2, we propose that APBs are able to induce compaction of telomeric chromatin and reduce TRF2 levels at these telomeres.

A mechanism that could lead to a reduced binding of TRF2 to the telomere repeats in APBs is post-translational modification of TRF2 by the SUMO E3 ligase MMS21 and PARP-2, which have both been found to be enriched in APBs (Dantzer et al., 2004; Potts and Yu, 2007). In line with a previous study (Potts and Yu, 2007), knockdown of these proteins reduced APB formation in our RNAi screen. The relevance of sumoylation of shelterin and PML-NB components for PML-NB and APB formation has been described in a number of previous studies (Brouwer et al., 2009; Chung et al., 2011; Hattersley et al., 2011; Lang et al., 2010; Potts and Yu, 2007; Yu et al., 2010). Here, we additionally found that MMS21 knockdown increased TRF2 binding to telomeres in APBs, whereas knockdown of the SUMO protease SENP6 resulted in a decrease. Thus, our results support the previous hypothesis that recruitment of MMS21 to APBs leads to shelterin destabilization at these telomeres, possibly by interfering with TRF2 dimerization (Potts and Yu, 2007). Interestingly, knockdown of PARP-2 also increased the ratio of the TRF2 signal to the telomere repeat signal. It is known that PARP-2 covalently modifies the dimerization domain of TRF2 and non-covalently binds poly(ADP-ribose) to the MYB domain of TRF2, which decreases the DNA-binding affinity of TRF2 (Dantzer et al., 2004). Thus, the enrichment of MMS21 and PARP-2 in APBs could reduce the level of TRF2 bound to telomeres in APBs by interfering with TRF2 dimerization and DNA binding.

Short-term TRF2 depletion has previously been shown to increase the rate of telomeric sister chromatid exchanges (T-SCEs) (Zeng et al., 2009). However, TRF2 is also important for t-loop formation and prevents homologous-recombination-induced t-loop deletions and chromosome fusions mediated by NHEJ (Wang et al., 2004). In addition, long-term depletion of TRF2 in ALT cells leads to chromosome fusions by NHEJ, induction of senescence and telomere shortening due to uncontrolled recombination (Stagno D'Alcontres et al., 2007). Thus, we hypothesize that ALT cells depend on partial telomere deprotection to drive telomere recombination. At the same time, they need to prevent an extensive loss of TRF2, which would lead to telomere attrition and chromosome fusions as discussed previously (Cesare and Reddel, 2010). Based on the results described here, we conclude that APBs induce the formation of this ‘intermediate-state’.

A previous report has shown that ATM is constitutively activated in ALT cells and colocalizes with APBs (Stagno D'Alcontres et al., 2007). Here, we show that ATM is preferentially activated in APBs that contain the densest telomere repeats. These highly dense telomere repeats had reduced levels of TRF2 bound per repeat. In addition, previous studies have found that TRF2 inhibits ATM by directly interacting with the region containing S1981, a residue whose autophosphorylation is necessary for the activation of this kinase (Denchi and de Lange, 2007; Karlseder et al., 2004). Accordingly, we propose that the reduction of TRF2 binding due to APB formation triggers ATM activation specifically at telomeres in APBs. The events subsequent to the DNA damage response, downstream of ATM-like recruitment of other DNA repair proteins and DNA repair synthesis (as detected by BrdU incorporation at APBs), have been addressed in our work that exploits ectopic APB assembly (Chung et al., 2011). Other studies have reported that multiple dysfunctional telomeres in ALT-positive cells colocalize with APB-like foci (Cesare et al., 2009) and that the phosphorylated histone H2AX (γH2AX), a molecular marker of double-strand breaks (DSBs) is found at some APBs (Nabetani et al., 2004). Here, we extended these observations by showing that PML knockdown reduced the number of telomeres colocalizing with p-ATM, whereas the total number of detectable p-ATM foci was increased (Fig. 4D). Thus, ATM was activated at telomeres after APBs were formed as opposed to a mechanism by which APBs assemble at telomeres where a DDR was already initiated. In addition, we found that inhibition of ATM did not affect the number of APBs, but decreased C-circle levels and reduced telomere repeat content, presumably due to a suppressed DDR at telomeres in APBs. In support of these results, it has been previously reported that ATM activity in ALT cells is not required for APB formation, but for telomeric DNA synthesis (Nabetani et al., 2004). Inhibiting the latter process in ALT cells does not have an immediate effect on cell viability and proliferation (Jiang et al., 2005; Potts and Yu, 2007). Consistent with this, ATM inhibition in U2OS cells hardly affects their survival on the time scale of several days, as is apparent from the experiments shown in Fig. 4 and in agreement with the findings from another study (Flynn et al., 2015).

In summary, our results demonstrate that PML induces compaction and confined TRF2 depletion at telomeres in APBs, and promotes telomere lengthening by initiating DNA damage signaling. Thus, APBs exert a central function in the disease phenotype of ALT-positive tumors.

For the inducible PML knockdown, a double-stranded DNA (dsDNA) oligonucleotide consisting of a microRNA (miRNA) against PML was cloned into the pcDNA6.2-GW/EmGFPmiR vector (Invitrogen). The complete miRNA and emerald green fluorescent protein (EmGFP) coding sequence were then cloned into the inducible pT-Rex-DEST30 vector (Invitrogen). Sequences of the dsDNA oligonucleotides for PML knockdown were: top oligonucleotide: 5′-TGCTGTCTTGGATACAGCTGCATCTTGTTTTGGCCACTGACTGACAAGATGCATGTATCCAAGA-3′; and bottom oligonucleotide, 5′-CCTGTCTTGGATACATGCATCTTGTCAGTCAGTGGCCAAAACAAGATGCAGCTGTATCCAAGAC-3′. The fluorescence three-hybrid system for recruiting GFP-tagged proteins to lacO arrays through GBP–LacI and GBP–LacI–RFP was used as described previously (Chung et al., 2011). The pEGFP-N2-mHMGN5 vector was kindly provided by Michael Bustin (Center for Cancer Research, National Cancer Institute, Bethesda, USA) (Rochman et al., 2009). The pEGFP-HP1γ plasmid was obtained by amplifying the human HP1γ cDNA sequence by PCR with an upstream forward primer, containing a BspEI restriction site, and a downstream reverse primer, containing a BamHI restriction site. The PCR product was then cloned in pEGFP-C1 (Clontech, Palo Alto, CA).

Human U2OS osteosarcoma cells (ATCC) and the U2OS cell clones with integrated lacO arrays, F6B2 and F42B8 (Jegou et al., 2009), were cultured in Dulbecco's modified Eagle's medium (DMEM; GIBCO) containing 10% fetal bovine serum (FBS; PAA) and 2 mM L-glutamine (PAA). The cell line stably expressing PML miRNA and EmGFP was constructed by co-transfection of the inducible pT-Rex-DEST30 vector containing a PML miRNA and EmGFP (Invitrogen) together with the Tet-repressor-coding vector pcDNA6/TR (Invitrogen). The selection was conducted with G418 and Blasticidin, and stable cell clones were picked and cultured for 10 days. The surviving cell clones were split into two fractions, and one fraction was maintained in doxycycline-free medium. For these cells, complete repression of the miRNA was ensured by analyzing GFP expression levels. The other fraction was induced with medium containing 1 µg/ml doxycycline (Sigma) for 24 h. The cell clone with the best repression in the uninduced state and best expression upon induction was used. The efficiency of PML knockdown was assessed by immunofluorescence against PML after 72 h of induction. For long-term PML knockdown, cells were cultured in medium containing 1 µg/ml doxycycline. Control cells were maintained in doxycycline-free medium. For the screening, 80,000 cells were seeded per slide on Lab-Tek chambered cover glasses (Thermo Scientific) and fixed after 72 h. For recruitment assays, cells were transfected using Effectene (Qiagen) according to the manufacturer's instructions and fixed after 24 h. For inhibition of histone deacetylases, cells were treated with 2 µM SAHA (Millipore) for 24 h and fixed afterwards. ATM was inhibited using 10 µM of the inhibitor KU-55933 (Hickson et al., 2004) (Calbiochem).

After fixation with 4% paraformaldehyde in PBS for 12 min and washing three times with PBS, cells were permeabilized for 5 min with 0.1% Triton X-100 in PBS. After three PBS washes, cells were blocked for 1 h with 10% goat serum in PBS and afterwards incubated with primary antibody in 10% goat serum in PBS for 1 h. Cells were then washed three times with PBS containing 0.002% (v/v) NP-40. Subsequent staining with the appropriate secondary antibodies conjugated to fluorescent dyes was conducted for 1 h in 10% goat serum in PBS. After washing the cells three times with PBS, cells were mounted with ProLong Gold (Invitrogen) containing 4ʹ,6-diamidino-2ʹ-phenylindole (DAPI). The following antibodies were used: mouse anti-TRF2 (1:100, 4A794, Calbiochem), mouse anti-ATM phosphorylated at S1981 (1:100, #MAB3806, Millipore), mouse anti-Cy3/Cy5 (1:500, #ab52060, Abcam), rabbit anti-phospho-H2A.X (Ser139) (1:100, #07-164, Millipore), rabbit anti-PML (1:100, #sc-5621, Santa Cruz Biotechnology), mouse anti-BrdU (1:50, B44, BD Biosciences), goat anti-mouse-IgG conjugated to Alexa Fluor 488 (1:300, Invitrogen), goat anti-mouse-IgG conjugated to Alexa Fluor 568 (1:300, Invitrogen), goat anti-rabbit-IgG conjugated to Alexa Fluor 488 (1:300, Invitrogen) and goat anti-rabbit-IgG conjugated to Alexa Fluor 633 (1:200, Invitrogen).

For 5-bromo-2’-deoxyuridine (BrdU) staining, cells were seeded, transfected and incubated for 2 days. After adding 100 µM BrdU (Sigma-Aldrich) to the medium for 2 h, cells were fixed and permeabilized with PBS containing 0.2% (v/v) Triton X-100. Cells were denatured with 1.5 M HCl for 30 min, blocked and stained with an antibody against BrdU as described above.

For telomere FISH, cells were washed three times with PBS and fixed with 4% paraformaldehyde in PBS for 12 min. After 5 min permeabilization with 0.2% (v/v) Triton X-100 in PBS, cells were dehydrated in a series of ethanol washes (70, 85 and 100% ethanol) for 2 min each. After air-drying, the samples were incubated with a Cy3-labeled (CCCTAA)₃ PNA probe (0.1 µM, Panagene Inc.) in 75% formamide in 20 mM NaCl, 20 mM Tris-HCl, 0.1% BSA, pH 7.4. Samples were denatured at 80°C for 3 min and hybridized overnight at 30°C. Slides were then washed consecutively with 70% formamide in 10 mM Tris-HCl pH 7.4, 2× SSC, 0.1× SSC at 55°C and 0.05% Tween-20 in 2× SSC (v/v). Subsequent immunofluorescence was conducted as described above. Quantitative FISH on metaphase spreads (Q-FISH) was performed as described previously (Poon and Lansdorp, 2001).

Confocal fluorescence images were acquired with a Leica TCS SP5 DMI6000 confocal laser scanning microscope (oil immersion objective lens, 63×, 1.4 NA). The automated screening was conducted as described previously (Osterwald et al., 2012). For manual image acquisition, images were acquired with the Leica TCS SP5 DMI6000 confocal laser scanning microscope using the LAS AF software and parameters as described above. The automated image analysis was performed using a 3D-model-based segmentation approach (Osterwald et al., 2012; Wörz et al., 2010).

The relative frequency distributions in Fig. 1D, Fig. 2A–C, Fig. 3A, B and Fig. 4F were obtained by binning the data and plotting the relative frequencies of telomeres or cells in each bin together with the corresponding s.e.m. as data points connected by lines. The analysis of metaphase telomere FISH was performed with the automated image analysis pipeline described above. Interphase cells and telomere repeat foci not associated with chromosomes (ECTRs) were excluded from the analysis. The manual analysis of microscopy images was performed with the ImageJ software (http://rsbweb.nih.gov/ij). For the analysis of the recruitment efficiency to lacO arrays, spots were counted as colocalizing if the signal at the lacO array was at least twofold above the background and comprised at least two pixels with a size of 200 nm.

The C-circle assay was performed as described previously (Henson et al., 2009). Briefly, DNA was isolated from 1×10⁶ cells using the QIAamp DNA Mini Kit (Qiagen). DNA was quantified using a Qubit Fluorometer (Life Technologies). Genomic DNA (20 ng) was digested with 12.5 U/μg HinfI and RsaI restriction enzymes (both Roche) and 5000 ng/μg RNase A (Thermo Fisher Scientific) for 2 h at 37°C. The digested DNA (10 µl) was combined with 10 µl 1× Φ29 Buffer, 7.5 U Φ29 DNA polymerase (both NEB), 0.2 mg/ml BSA, 0.1% (v/v) Tween 20, 1 mM each dATP, dGTP and dTTP and incubated for 8 h at 30°C and then at 65°C for 20 min. After adding 40 µl 2× SSC, the sample was dot-blotted onto a 2×-SSC-soaked Roti-Nylon plus membrane (pore size 0.45 µm, Carl Roth). The membrane was baked for 20 min at 120°C and hybridized and developed using the TeloTAGGG Telomere Length Assay Kit (Roche). Intensities of C-circle dot blots were analyzed and background-corrected using Image Lab 4.1 (Bio-Rad).

Genomic DNA was purified using the Gentra Puregene Cell Kit (Qiagen). For terminal restriction fragment (TRF) analysis, 5 µg of purified DNA was digested with HinfI and RsaI overnight. The digested DNA was resolved on a 0.6% agarose gel (Biozym Gold Agarose) in 1× TAE buffer using the CHEF-DRII pulsed-field gel electrophoresis system (Bio-Rad) with the following settings: 4 V/cm, initial switch time 1 s, final switch time 6 s, and 13 h duration. Southern blotting and chemiluminescent detection was performed using the TeloTAGGG Telomere Length Assay Kit (Roche) according to the manufacturer's instructions. The blot was visualized with a ChemiDoc MP imaging system (Bio-Rad). Approximate mean TRF lengths were quantified using ImageJ and according to the following equation: Σ(OD_i)/Σ(OD_i/L_i), where OD_i is the optical density at position i and L_i is the TRF length at position i.

Telomere-repeat quantitative PCR was conducted essentially as described previously (Cawthon, 2002; O'Callaghan et al., 2008). In short, 5 or 10 ng DNA, 1× LightCycler 480 SYBR Green I Master (Roche), 500 nM forward primer and 500 nM reverse primer were added per 10 μl reaction. The primer sequences were: telo fwd, 5′-CGGTTTGTTTGGGTTTGGGTTTGGGTTTGGGTTTGGGTT-3ʹ; and telo rev, 5′-GGCTTGCCTTACCCTTACCCTTACCCTTACCCTTACCCT-3ʹ; 36B4 fwd, 5′-AGCAAGTGGGAAGGTGTAATCC-3ʹ; and 36B4 rev, 5′-CCCATTCTATCATCAACGGGTACAA-3ʹ. Cycling conditions (for both telomere and 36B4 products) were 10 min at 95°C, followed by 40 cycles of 95°C for 15 s and 60°C for 1 min. A standard curve was used to determine relative quantities of telomere repeats (T) to that of a single copy gene (S, 36B4 gene, also known as RPLP0). The T:S ratio was calculated and normalized to a reference T:S ratio.

Transfected cell microarrays were produced as previously described (Erfle et al., 2007). Repetitions of a 4×4 array were printed on each Lab-Tek resulting in 384 spots with 24 replicates for each siRNA. A gene was considered as a hit if knockdown with two different siRNAs consistently affected the number of APBs by more than 10% (P<0.05). The experiments were conducted in triplicates, and 500 to 1500 cells were analyzed per siRNA. Sequences of all siRNAs (silencer select siRNAs, Ambion) as well as reported knockdown efficiencies, if available, can be found in supplementary material Table S1. The knockdown efficiencies of selected siRNAs that were important for further conclusions, namely CBX3, HDAC7, MRE11, NBS1, NSBP1, PARP2, RAD50, RAP1A, SENP6, SUV420H2 and TRF2, were validated by real-time quantitative PCR (supplementary material Fig. S3A). Values were normalized against β-actin. Primer sequences are provided in supplementary material Table S3. GO terms for biological processes associated with hits were identified by using the gene ontology website (http://geneontology.org).

To analyze the effect of each siRNA on the cell cycle, the background-corrected integrated DAPI intensities that were obtained in the automated high-content confocal screening were normalized and histograms were plotted (supplementary material Fig. S3B). As described previously, the distributions obtained in this manner correlate well with fluorescence-activated cell sorting (FACS) profiles (Tóth et al., 2004; Osterwald et al., 2012). Gates (e.g. for minimum and maximum DAPI intensity thresholds) were defined to obtain and compare the relative percentage of cells in G1, S and G2/M phase for each siRNA transfection. The same binning and gating was used for all conditions. The percentage of cells in each cell cycle phase was obtained from at least three replicates for each siRNA transfection. These data were used to calculate changes in the percentage of cells in G1, S and G2/M phase induced by each siRNA relative to control siRNA and the corresponding s.e.m.

The statistical analysis was conducted using the R software (http://www.r-project.org) as described previously (Osterwald et al., 2012). Errors bars always represent the s.e.m. of at least three independent experiments, unless stated otherwise. A Kolmogorov–Smirnov test was used to assess the significance of siRNA-related effects (supplementary material Table S2) and for the evaluation of interphase and metaphase FISH results with respect to changes in telomere repeat intensities or densities as well as TRF2:telomere repeat ratios. Welch's t-test was applied for the analysis of changes in cell cycle distribution, C-circle levels, the number of telomere-free ends, ECTRs and the percentage of ECTR intensity of total telomere intensity. For the analysis of the recruitment efficiency to lacO arrays, the percentage of lacO arrays with colocalization was determined with the indicated value n being the number of lacO arrays evaluated. Error bars were calculated as n, which yields the standard deviation for a Poisson distribution. In order to determine whether the percentages of colocalization after recruiting the proteins of interest were significantly different from the ones obtained in the controls, the two-sided Fisher's exact test was used to calculate P-values.

We thank Nina Beil, Fabian Erdel, Delia Braun, Jürgen Reymann, Jana Molitor, Jan-Philipp Mallm and Brian Luke for help and discussions, and Michael Bustin for plasmid vectors.