Biphasic recruitment of TRF2 to DNA damage sites promotes non-sister chromatid homologous recombination repair

TRF2 (also known as TERF2) is an integral component of the telomere shelterin complex that protects telomere integrity (Bilaud et al., 1997; Feuerhahn et al., 2015; Okamoto et al., 2013; van Steensel et al., 1998). TRF2 recognizes the telomere repeat sequence directly through its C-terminal MYB/SANT domain, and protects telomeres by both promoting T-loop formation and inhibiting the DNA damage checkpoint kinase, ATM (de Lange, 2002; Griffith et al., 1999; Karlseder et al., 2004; van Steensel et al., 1998). TRF2 has also been shown to be recruited to non-telomeric DNA damage sites and promote DNA double-strand break (DSB) repair, though its exact role in the process remains controversial. Recruitment of TRF2 has been observed at high-irradiance laser-induced DNA lesions, but not at damage sites induced by low-irradiance ultraviolet radiation or ionizing radiation, despite the presence of DSBs in both cases (Bradshaw et al., 2005; Huda et al., 2012; Williams et al., 2007). TRF2 has also been linked to homologous recombination (HR) repair (Mao et al., 2007), but its phosphorylation by ATM appears to be important for fast repair [suggested to be non-homologous end-joining (NHEJ)] (Huda et al., 2009). Thus, TRF2 recruitment and function at non-telomeric DNA damage sites remain enigmatic.

PARP1 is a DNA nick sensor activated rapidly and transiently in response to DNA damage (for reviews see Ball and Yokomori, 2011; Beck et al., 2014; Daniels et al., 2015; Kalisch et al., 2012). There are multiple PARP family members, but PARP1 plays a major role in poly(ADP-ribose) (PAR) response at damage sites (Cruz et al., 2015; Kong et al., 2011). Activated PARP1 uses NAD⁺ as a substrate to ADP-ribosylate multiple target proteins, including itself. Although PARP1 was initially thought to specifically facilitate base excision repair and/or single-strand break repair, recent studies have revealed its role in multiple DNA repair pathways, including DSB repair (Beck et al., 2014). Furthermore, PAR modification at DNA damage sites is critical for the recruitment of chromatin modifying enzymes that promote DNA repair (Ahel et al., 2009; Ayrapetov et al., 2014; Ball and Yokomori, 2011; Chou et al., 2010; Gottschalk et al., 2009; Izhar et al., 2015; Khoury-Haddad et al., 2014; Larsen et al., 2010; Polo et al., 2010; Smeenk et al., 2010; Sun et al., 2009). Thus, PARP1 is not only a sensor of DNA damage, but also a regulator of damage site chromatin environment and multiple DNA repair pathways in higher eukaryotes.

Previously, we found that TRF2 is rapidly recruited to higher input-power laser damage sites that contained complex DNA damage in a PARP-dependent manner (Cruz et al., 2015). In contrast, lower input-power laser irradiation that induced relatively simple strand-breaks and no significant PAR response failed to recruit TRF2 (for laser power measurements, see Materials and Methods). Importantly, stimulation of PARylation at lower input-power laser damage sites by a PARG inhibitor promotes TRF2 recruitment in HeLa cells (Fig. 1A). Therefore, the level of PARP activation, rather than the nature of damage per se, is the deciding factor for TRF2 recruitment to non-telomeric DNA damage sites.

We and others have observed rapid and transient TRF2 recruitment to damage sites within the first 5 min post-irradiation (p.i.) (Bradshaw et al., 2005; Cruz et al., 2015) (Fig. 1). Upon inspection of later time points (20–30 min), however, we found that TRF2 re-appears at damage sites (Fig. 1B,C). Similar recruitment patterns were observed with both the ectopically expressed and endogenous TRF2 (Fig. 1B,D, respectively). The initial recruitment of GFP–TRF2 peaks at ∼1–3 min p.i. (termed ‘phase I’), which decreases once but returns peaking at ∼30 min to 1 h (‘phase II’) (Fig. 1A–C). This phase II recruitment persists for at least 2 h (data not shown). Interestingly, the PARP inhibitors NU1025 and olaparib completely suppressed phase I, but not phase II, recruitment of both endogenous and recombinant TRF2 (Fig. 1D,E), suggesting that two phases of TRF2 recruitment are mediated by different mechanisms.

We found that transient GFP–TRF2 displacement is inversely correlated with the appearance of a prominent dark line at damaged lesions readily visible using bright-field microscope imaging (Fig. 1B,F). Close examination at damage sites revealed that GFP signals not only decrease, but are often transiently pushed to the periphery of the damage sites, suggesting that it may be displaced by the constituents of the dark line (Fig. 1B,F). Intrinsically disordered proteins (IDPs) FUS, EWS and TAF15 (collectively, FET), have been shown to accumulate at damage sites in a PAR-dependent manner and are the major components of the dark line (Altmeyer et al., 2015). We found that depletion of FET by siRNA resulted in a more even distribution of the GFP–TRF2 signal at damaged lesions, correlating with disappearance of the dark line (Fig. 1F; Fig. S1A). The results reveal that although both TRF2 and IDPs are recruited to damage sites in a PAR-dependent manner, there is a distinct order of appearance and competition between them, which separates phases I and II.

TRF2 protein domains have been characterized extensively in the context of telomeres. The C-terminal MYB/SANT DNA binding domain of TRF2 specifically recognizes and binds telomere DNA whereas the N-terminal basic domain is not required for telomere targeting (Fig. 2A) (Karlseder et al., 1999; Okamoto et al., 2013). We found that phase I recruitment is abolished by deletion of the N-terminal basic domain, with results similar to those previously reported by Bradshaw et al. (2005) (Fig. 2B,C). In contrast, phase II recruitment was significantly inhibited by deletion of the C-terminal MYB/SANT domain. Deletion of both N- and C-terminal domains abolished both phases of damage site recruitment. Thus, distinct domains play critical roles in rapid and transient phase I and slow and stable phase II. Unlike the previous report (Bradshaw et al., 2005), we found that deletion of the MYB/SANT domain partially reduced phase I recruitment, suggesting that the phase I recruitment is further stabilized by DNA binding of the MYB/SANT domain (Fig. 2C).

The N-terminal basic domain harbors multiple arginine residues, which may interact with negatively charged PAR residues clustered at damage sites. Indeed, arginine-to-alanine mutations completely abolished the phase I recruitment (Fig. 2D, RA). In contrast, this recruitment is sustained, albeit more weakly, by arginine-to-lysine mutations that preserve the positive charge (Fig. 2D, RK). The basic domain was also shown to bind to the Holliday junction (HJ) (Fouché et al., 2006). However, a HJ binding-defective mutation (H31A) (Poulet et al., 2009) showed no inhibitory effect on TRF2 recruitment to damage sites (Fig. 2D). Thus, the positive charge is essential for the PARP-dependent TRF2 recruitment to non-telomeric DNA damage sites.

Human telomerase reverse transcriptase (TERT) is responsible for the addition of telomere sequences. Several studies have indicated that it can also polymerize DNA at non-telomeric DNA ends de novo and has a distinct role in DNA damage response (DDR) and repair (Flint et al., 1994; Gao et al., 2008; Majerská et al., 2011; Masutomi et al., 2005; Morin, 1991; Ribeyre and Shore, 2013). Because the MYB/SANT domain, which is responsible for telomere repeat recognition, is pertinent to phase II recruitment, we tested the possible contribution of TERT to TRF2 association at DSB sites. HeLa cells, which express TERT, were treated with siRNA specific for TERT (Fig. 3A; Fig. S2A). We found that the phase II recruitment of TRF2 was partially inhibited by this treatment (Fig. 3A). TRF2, however, can also be recruited to damage sites even in telomerase-negative alternative lengthening of telomeres (ALT) cells, albeit at a lower level (e.g. U2OS cells) (Fig. S2B). Thus, the results indicate that TERT contributes to, but is not essential for, MYB/SANT domain-dependent TRF2 recruitment to damage sites.

Chimeric mutants combining the domains of TRF1 (also known as TERF1) and TRF2 have provided important insight into the specific role of TRF2 in telomere protection (Okamoto et al., 2013). The N-terminal basic domain is unique to TRF2 (TRF1 has an acidic domain), indicating that phase I recruitment is specific to TRF2. Similarly, we found that the TRF2 TRFH domain required for dimerization (<30% homology with TRF1) is essential for both phase I and II recruitment (Fig. 3B,C; TRFcT). Interestingly, the TRFcH mutant replacing the TRF2 hinge domain with that of TRF1 exhibited intact phase I recruitment, but failed for phase II, indicating that the hinge domain (11% homology to TRF1) is required for phase II recruitment. In contrast, the TRF2 MYB/SANT domain was found to be interchangeable with that for TRF1 for phase II recruitment (both recognize telomere repeats) (Fig. 3B,C; TRFcM).

The hinge domain contains binding sites critical for the interaction between TRF2 and several different factors (Chen et al., 2008; Okamoto et al., 2013). Deletion of the TIN2 binding region (ΔTIN2; amino acids 352–367) critical for TRF2 incorporation into the shelterin complex (Kim et al., 2004; Liu et al., 2004; Ye et al., 2004) had no effect on damage site recruitment, suggesting that TRF2 recruitment to damage sites is independent of the shelterin complex (Fig. 3D). The hinge domain also contains a region critical for suppression of DDR and telomere maintenance, termed ‘inhibitor of DDR’ (iDDR; amino acids 406–432) (Okamoto et al., 2013). Deletion of this domain (ΔiDDR) recapitulated the phenotype of the TRFcH mutant, inhibiting only the phase II recruitment (Fig. 3D). The iDDR region has previously been shown to be necessary and sufficient for the interaction of TRF2 with the MRE11 complex (Okamoto et al., 2013). We found that siRNA depletion of MRE11 and NBS1 (also known as NBN), the two components of the complex, effectively reduced phase II recruitment (Fig. 3E; Fig. S1B). These results indicate that the MYB/SANT domain-dependent phase II requires the interaction of the iDDR region with the MRE11 complex.

I-PpoI endonuclease induces DSB sites in the ribosomal DNA region (Berkovich et al., 2008). Using chromatin immunoprecipitation (ChIP)-qPCR, we found binding of TRF2 to these DSB sites (Fig. 3F). Intriguingly, this binding was further enhanced by methyl methanesulfonate (MMS) treatment, which induces complex damage and stronger PARP activation than simple DSBs. Importantly, TRF2 accumulation at I-PpoI target sites both in the presence and absence of MMS treatment was effectively suppressed by PARP inhibitor, confirming that TRF2 binding to DSB sites is PARP-dependent (Fig. 3F). Interestingly, depletion of the MRE11 complex also abolished TRF2 binding to DSB sites. These results indicate that strong PARP activation is the key determinant for efficient TRF2 recruitment to DSB sites and that both phase I and II recruitment mechanisms are important for the stable binding of TRF2 to DSB sites.

To determine the significance of TRF2 recruitment to damage sites, the effects of TRF2 depletion were examined using cell-based assays for different pathways of DSB repair (Hu and Parvin, 2014). TRF2 was depleted for 48 h before repair assays in order to minimize telomere erosion, which was typically assayed 4–7 days after depletion (Okamoto et al., 2013; Rai et al., 2016) (Fig. S3). TRF2 has previously been implicated in both HR repair (Mao et al., 2007) and fast repair (suggestive of NHEJ) (Huda et al., 2009). Interestingly, we found that TRF2 depletion reduced the efficiency of HR in the I-SceI-dependent HR assay, but not in the sister chromatid exchange (SCE) assay (Fig. 4A,B). The I-SceI assay selectively captures intra-chromatid or unequal sister chromatid HR, whereas the SCE assay specifically detects sister chromatid conversion (Potts and Yu, 2005). Furthermore, TRF2 depletion had no significant effect on either classical or alternative NHEJ repair (Fig. 4B). Thus, the results indicate that TRF2 plays a specific role in non-sister chromatid HR. Under our experimental conditions, the effect of TRF2 depletion on I-SceI HR was comparable to that of BRCA1 depletion and can be complemented by expression of wild-type TRF2 (Fig. 4A; Fig. S1C). BRCA1 is a known promoter of HR (Moynahan et al., 1999). Although the N- or C-terminal deletion mutants of TRF2, which primarily affect phase I or phase II, respectively, exhibited variable results, the TRFcT mutant that hinders both phases clearly failed to complement, indicating that both phases are critical for its optimal repair activity (Fig. 2C, Fig. 3C, Fig. 4). This is consistent with the requirement of both phases for stable TRF2 binding at DSB sites (Fig. 3F).

The selective involvement of TRF2 in non-sister chromatid HR is an interesting contrast to cohesin, which only promotes sister chromatid HR but not other types of HR (Kong et al., 2014; Potts et al., 2006). These differences can be explained by their mechanisms of actions. Cohesin promotes sister chromatid cohesion, therefore promoting pairing of damaged and undamaged sister chromatids for HR. TRF2 promotes loop formation and strand invasion not only at telomere T-loops, but also with non-telomeric templates in vitro (Fig. 4C) (Amiard et al., 2007; Doksani et al., 2013; Griffith et al., 1999; Stansel et al., 2001). Furthermore, TRF2 has been shown to inhibit Rad51-mediated D-loop formation with a telomeric, but not non-telomeric, template in vitro (Bower and Griffith, 2014). Thus, the cell may hijack the function of TRF2 by clustering the protein to DSB sites to promote intra-chromatid HR, which may be particularly important in the context of complex damage that robustly activates the PARP response.

Our results demonstrate that the initial phase I recruitment by PARP may be critical to bring TRF2 to the proximity of non-telomeric damage sites, which is then further stabilized by the Mre11 complex interaction and DNA binding through the MYB/SANT domain (Fig. 4D). PAR-dependent accumulation of IDPs may compete with phase I recruitment and trigger the second mode of TRF2 association with DNA lesions. Thus, PARP initiates the cascade of dynamic recruitment of factors, some of which compete with each other, to fine-tune repair pathway choice. Interestingly, a recent study showed that BLM is also recruited to damage sites in a biphasic fashion, initially by ATM signaling, and subsequently by the MRE11 complex (Tripathi et al., 2018), suggesting that this type of two-step mechanism may be commonly used to ensure versatility and specificity of factor recruitment.

Our results demonstrate PARP- and MRE11 complex-dependent recruitment of TRF2 to DSB sites, reconciling previous controversies regarding TRF2 recruitment and function at non-telomeric DNA damage sites and revealing a uniquely regulated non-telomeric and shelterin-independent function of TRF2 in non-sister chromatid HR repair.

HeLa and U2OS cells (American Type Culture Collection) were grown in Dulbecco's modified Eagle's medium (DMEM; Gibco) supplemented with L-glutamine, 10% fetal bovine serum (FBS) and antibiotics. HeLa DR–GFP (for HR assay) (Hu and Parvin, 2014; Pierce et al., 2001) and EJ2–GFP (for alternative NHEJ assay) (Bennardo et al., 2008) stable cells were grown in DMEM with high glucose (4500 mg/l), supplemented with 10% (v/v) FBS (Omega Scientific), 1% pen/strep, 1% GlutaMAX (both Gibco) and 1.5 μg/ml puromycin (Sigma). 293/HW1 cells (for classic NHEJ assay) (Zhuang et al., 2009) were grown in DMEM (high glucose), 1% pen/strep, 1% sodium pyruvate, 1% GlutaMAX (all Gibco), 10% FBS (Omega Scientific), and 2 μg/ml puromycin (Sigma). All the cell lines were mycoplasma-negative.

The expression plasmids containing GFP–TRF2 full-length, deletion or TRF1/2 chimeric mutants were kindly provided by Dr Eros Lazzerini Denchi (The Scripps Research Institute, La Jolla, CA, USA) (Okamoto et al., 2013). These cDNAs were also re-cloned into a pIRES vector containing an N-terminal FLAG epitope. Those include TRF2ΔN (amino acids 45–500), TRF2ΔC (amino acids 1–454), TRF2ΔNΔC (amino acids 45–454), ΔTIN2, and TRFcT (Okamoto et al., 2013). RA and RK mutants were as described previously (Mitchell et al., 2009).

Mouse monoclonal antibodies specific for PAR polymers [BML-SA216–0100, Enzo Life Sciences; 1:500 dilution for immunofluorescence staining (IF)], TRF2 (NB100–56506, Novus Biologicals; 1:500 for IF), MRE11 [GTX70212, GeneTex; 1:500 for IF, 1:1000 for western blotting (WB)], GFP (632592, Takara Bio; 1:500 for ChIP), Actin (A4700, Sigma; 1:1000 for WB), FLAG (F3165, Sigma; 1:1000 for WB), and BRCA1 (GTX70111, GeneTex; 1:1000 for WB), as well as rabbit polyclonal antibodies specific for γH2AX (GTX628789, GeneTex; 1:500 for IF), EWSR1 (GTX114069, GeneTex; 1:1000 for WB), TAF15 (GTX103116, GeneTex; 1:1000 for WB), H3 (14-411, Upstate Bio; 1:1000 for WB) and Rad21 (Kong et al., 2014; 1:2000 for WB) were used.

At different time points after damage induction, cells were fixed in 4% paraformaldehyde (15 min at 4°C), permeabilized in 0.5% Triton X-100 for 5 min (4°C), and stained with antibodies. The staining procedure was as described previously (Kim et al., 2002). Fluorescence images were captured through a 100× Ph3 UPlanFI oil objective lens (NA, 1.3; Olympus) on a model IX81 Olympus microscope with a charge-coupled device camera.

Protein samples were subjected to SDS-PAGE and then transferred to nitrocellulose membranes as described previously (Schmiesing et al., 1998). The membranes were blocked with Pierce Protein-Free T20 (PBS) Blocking Buffer (Thermo Fisher Scientific). The primary antibody was incubated in 3% BSA+0.05% Tween 20 in PBS for 1 h at room temperature or overnight at 4C°, followed by three washes in PBS+0.05% Tween 20. The secondary antibody conjugated to HRP (anti-mouse-IgG, W4021; anti-rabbit-IgGm W4011, both used at 1:10,000; Promega) was incubated in 3% BSA+0.05% Tween 20 in PBS for 1 h at room temperature. The filter was then washed three times in PBS+0.05% Tween 20 and developed with SuperSignal West Pico Chemiluminescent Substrate (Thermo Fisher Scientific). Images were acquired using the Image Analyzer (LAS-4000, Fujifilm) and analyzed using Quantity One software (Bio-Rad).

Near-infrared (NIR) femtosecond laser irradiation was carried out using a Zeiss LSM 510 META multiphoton-equipped (3.0 W 170 fs coherent tunable Chameleon Ultra-NIR laser) confocal microscope. The Chameleon Ultra-NIR beam was tuned to 780 nm, where the software bleach function was used to target linear tracts inside the cell nuclei for exposure to single laser scans (12.8 μs pixel dwell time) through the 100× objective lens (1.3 NA Zeiss Plan APO) (Cruz et al., 2015). The peak irradiance at the focal point for the higher input laser power is 5.27×10¹⁰ W/cm², and for the lower input laser power is 3.24×10¹⁰ W/cm² (Cruz et al., 2015). Fluorescence measurement of the recruitment of GFP–TRF2 wild type and mutant proteins, and other GFP-tagged proteins, to damage sites was analyzed using live-cell confocal scanning with the 488 nm CW argon laser on the Zeiss LSM 510 META platform. The signals were measured with the LSM 510 software (version 4.0). The data were collected in three separate experiments. P-values were calculated by two-tailed t-test using Excel software (Microsoft). Boxplots were created with R program (https://www.r-project.org).

HeLa cells were transfected twice, 24 h apart, with siRNAs at a final concentration of 5 nM using HiPerFect (Qiagen) transfection reagent according to the manufacturer's instructions. siRNAs directed against TERT (5′-TTTCATCAGCAAGTTTGGA-3′; Masutomi et al., 2005), MRE11 (5′-GCTAATGACTCTGATGATA-3′; Myers and Cortez, 2006), NBS1 (5′-GAAGAAACGTGAACTCAAG-3′; Myers and Cortez, 2006), human TRF2 (SI00742630, Qiagen), BRCA1 (5′-GCTCCTCTCACTCTTCAGT-3′; Hu and Parvin, 2014), FET [FUS (s5402), EWSR1 (s4886) and TAF15 (s15656); Ambion/LifeTechnologies; Altmeyer et al., 2015], and a negative-control siRNA (1027310, Qiagen) were used. Cells were harvested for western blot analyses or were subjected to laser micro-irradiation, approximately 48 h after the final transfection.

20 μM PARP inhibitor olaparib (Apexbio Technology) or 1 μM PARG inhibitor (PARGi) 6,9-diamino-2-ethoxyacridine lactate monohydrate (DEA; Trevigen) was added to the cell culture 1 h prior to damage induction. DMSO only was added to control cells.

The experimental procedure was previously described (Kong et al., 2014). Briefly, HeLa cells were transfected with expression plasmids encoding GFP–TRF2 (or GFP only) with or without ER-I-PpoI expression plasmid. 24 h later, 4-hydroxy-tamoxifen (4-OHT; Sigma) was added to a final concentration of 1 μM for 10 h to induce DNA damage. For MMS-treated cells, 3 mM MMS was added 1 h before harvest. DMSO or PARP inhibitor was added 1 h before MMS treatment. For the MRE11 and NBS1 depletion experiment, two rounds of control or MRE11 siRNA and NBS1 siRNA transfection were performed 48 h and 24 h before GFP–TRF2 and/or ER-I-PpoI plasmid transfection. ChIP analysis was performed, using GFP antibody, as described previously (Berkovich et al., 2008; Kong et al., 2014). The data represent the means±s.d. of two separate experiments. Four samples (absence or presence of I-PpoI, and with and without MMS) were repeated the third time, and results with the same trend were obtained. P-values were calculated by two-tailed t-test.

Homologous recombination (HR) and alternative NHEJ assays were performed as described previously in HeLa cells (Bennardo et al., 2008; Hu and Parvin, 2014) with modification. Briefly, on day 1, the appropriate cell lines [stable cell lines expressing appropriate repair templates; HeLa DR–GFP for the HR assay (Hu and Parvin, 2014; Pierce et al., 2001); EJ2–GFP for the alternative NHEJ assay (Bennardo et al., 2008), and 293/HW1 for the classic NHEJ assay (Zhuang et al., 2009)] were seeded in 24-well plates. The next day, cells at 50% confluence were transfected with TRF2 siRNA or control siRNA. On day 3, cells were re-transfected with the same siRNA for 5–6 h, and then were transferred to 35 mm wells. On day 4, the plasmid encoding I-SceI endonuclease was co-transfected with mCherry-expressing plasmid to induce DSBs. The cells were examined using flow cytometry on day 7, and the ratio of the GFP reporter to mCherry was used as a measure of HR or alternative NHEJ efficiency. The classic NHEJ assay utilizes real-time qPCR and was carried out as described (Hu and Parvin, 2014; Zhuang et al., 2009) in 293 cells with modification. The transfection procedure was as described above. For plasmid add-back in the rescue assays, the transfection procedure was the same except that at the second siRNA transfection, the blank control plasmid, and plasmids encoding FLAG–TRF2 and corresponding mutants were co-transfected into the cells using Lipofectamine 3000 (Thermo Fisher Scientific). The data represent the means±s.d. of three separate experiments. P-values were calculated by two-tailed t-test.

We thank Dr Jeff Parvin (Department of Biomedical Informatics, The Ohio State University, Columbus, OH) for kindly providing DSB repair assay cell lines, Dr Eros Lazzerini Denchi (The Scripps Research Institute, La Jolla, CA) for TRF mutant plasmids, and Dr Feng Qiao (Department of Biological Chemistry, School of Medicine, University of California, Irvine, CA) for helpful discussion and advice.