Characterization of t-loop formation by TRF2

ABSTRACT T-loops are thought to hide telomeres from DNA damage signaling and DSB repair pathways. T-loop formation requires the shelterin component TRF2, which represses ATM signaling and NHEJ. Here we establish that TRF2 alone, in the absence of other shelterin proteins can form t-loops. Mouse and human cells contain two isoforms of TRF2, one of which is uncharacterized. We show that both isoforms protect telomeres and form t-loops. The isoforms are not cell cycle regulated and t-loops are present in G1, S, and G2. Using the DNA wrapping deficient TRF2 Topless mutant, we confirm its inability to form t-loops and repress ATM. However, since the mutant is also defective in repression of NHEJ and telomeric localization, the role of topological changes in telomere protection remains unclear. Finally, we show that Rad51 does not affect t-loop frequencies or telomere protection. Therefore, alternative models for how TRF2 forms t-loops should be explored.


Introduction
Telomeres prevent the inappropriate engagement of DNA damage response pathways at the natural ends of linear chromosomes. The telomeric DNA in mammalian cells is associated with a telomerespecific six-subunit protein complex, shelterin, that represses at least six distinct DNA damage response pathways: ATM signaling, ATR signaling, PARP1 activation, Homology Directed Repair (HDR), NHEJ, and alternative Non-Homologous End Joining (alt-NHEJ) [1]. Shelterin is highly compartmentalized with different components of the complex functioning to repress distinct pathways.
TRF2 has been proposed to repress ATM signaling and NHEJ by remodeling telomeres into the t-loop structure wherein the ss 3 telomeric end is embedded in the duplex part of the telomere [19,20]. This sequestered telomere terminus is thought to be impervious to the MRN (Mre11/ Rad50/Nbs1) DSB sensor in the ATM kinase pathway and prevent the loading of the Ku70/80 initiator of NHEJ. Super-resolution imaging of t-loop frequencies in mouse embryo fibroblasts (MEFs) with and without TRF2 showed that t-loop formation strictly requires TRF2 function [19]. On the other hand, MEFs lacking the mouse POT1 proteins (POT1a and POT1b), Rap1, or TRF1 contain t-loops at normal frequencies [19]. This correlation between t-loop formation by TRF2 and the ability of TRF2 to repress ATM signaling and NHEJ underlies the model that the t-loop structure is a major impediment to activation of these pathways at telomeres. Nonetheless, TRF2 appears to have additional mechanisms to prevent NHEJ that involve its iDDR domain and its interaction with Rap1 [21][22][23].
While TRF2 was shown to be required for t-loop formation, it has not been established whether TRF2 is sufficient for the remodeling of telomeres. Conceivably, t-loop formation could involve a collaborative effort of TRF2 together with either TIN2 or TPP1 (which have not been tested) or TRF2 could cooperate with multiple shelterin components in a redundant fashion.
The mechanism by which TRF2 remodels telomeres into t-loops is of obvious interest. One model has been proposed based on the finding that the large dimerization domain of TRF2 referred to as the TRFH domain [3,24], weakly binds DNA and can wrap 90 bp around itself [25]. Topological stress induced by DNA wrapping could lead to DNA unwinding which is expected to promote the strand-invasion of the telomere 3 overhang. The wrapping activity of TRF2 requires the presence of seven lysine and two arginine residues on the surface of the TRFH domain [25]. Mutation of these residues to alanine (resulting in the 'Topless' mutant of TRF2) abrogates the DNA wrapping activity. TRF2 Topless was shown to be deficient in t-loop formation when expressed in human cells from which the endogenous TRF2 was depleted with shRNA, and under this condition ATM signaling was observed at a subset of the telomeres [25]. Importantly, NHEJ-mediated telomere fusions were not observed, unless Rap1 was removed [25]. These data nominated Topless TRF2 as a separation-of-function mutant and supported the topological model for t-loop formation and repression of ATM kinase activity. Given that these experiments were performed in cells that likely contained residual endogenous TRF2, we sought to further verify the results in a system that allows deletion of TRF2.
We present evidence that the Topless version of TRF2 is deficient in all aspects of telomere protection, suggesting that this is a loss-of-function mutant and therefore not readily interpretable. We show that telomeres containing TRF2 but not the other components of shelterin are present in the t-loop configuration and that both isoforms of TRF2 are equivalent in telomere protection. We also probe the effect of cell cycle stage on t-loop frequencies and examine t-loop formation in absence of Rad51, which has been implicated in t-loop formation using in vitro experiments designed to mimic this process [26]. The results indicate that neither cell cycle aspects nor Rad51 affect the frequency of t-loops at telomeres.

TRF2 is both necessary and sufficient for t-loop formation
In order to test whether TRF2 can form t-loops in the absence of other shelterin components, we made use of TRF1 F/F TRF2 F/F mouse embryo fibroblasts (MEFs) from which both TRF1 and TRF2 can be deleted by the Cre recombinase [27]. When these cells are provided with exogenous TRF2, the resulting telomeres contain TRF2, Rap1, TIN2, TPP1, and POT1 but no TRF1 [28]. In order to create conditions where the telomeres only contain TRF2, we complemented the cells with an exogenous version of TRF2 that contains deletions of its TIN2 and Rap1 bindings sites (TRF2ΔTIN2ΔRap1 or TRF2ΔTR) [29,30]. Under these conditions, neither TRF1 nor TIN2/TPP1/POT1 or Rap1 are present at the telomeres [27,28] (Figure 1(a)).
In order to determine whether t-loops are formed at telomeres containing only TRF2, we expressed wild type TRF2 and TRF2ΔTR in TRF1 F/F TRF2 F/F Lig4 -/-p53 -/-Cre-ER T2 MEFs. The use of cells deficient in DNA ligase IV avoids the formation of telomere fusions that confound t-loop analysis [19]. Wild type TRF2 and the TRF2ΔTR were expressed equally before and after Cre treatment and both versions of TRF2 were present in excess over the endogenous TRF2 (Figure 1(b)).
T-loop frequencies were determined using psoralen-UV cross-linked nuclei and the chromatin spreading technique previously established [19] and OMX-imaging of telomeric DNA labeled by FISH [31] (Figure 1(c)). Cells lacking TRF2 (vector control) showed background levels of t-loops, consistent with previous data [19] (Figure 1(c,d)). This t-loop defect was rescued by the expression of wild Figure 1. TRF2 is sufficient for the formation of t-loops. (a) Schematic of the experimental approach to create telomeres that contain TRF2 but no other shelterin components. (b) Immunoblot for TRF2 using the indicated MEFs with and without Cre treatment and expressing the indicated versions of TRF2 (wild type, wt; TRF2ΔTR, ΔTR; empty vector, vec). * nonspecific band; ctrl: nonspecific band used as loading control. Note that the endogenous TRF2 is not detected at this exposure. (c) Examples of the structure of telomeric DNA detected in the indicated MEFs treated with Cre (120 h) and infected as in (b). (d) Quantification of t-loop frequencies detected as in (c). Averages and SDs from three independent experiments (>100 molecules scored each). Significance determined using a two-tailed unpaired t-test (n.s. not significant; * p < 0.05; ** p < 0.01). (e) Immunoblots for TRF2 using the indicated MEFs infected with vector, TRF2, and TRF2ΔTR after Cre treatment. * nonspecific band. f, Examples of metaphase spreads with telomeres detected type TRF2, which is consistent with previous data showing that telomeres lacking TRF1 have normal t-loop levels [19]. The frequency of t-loops achieved by complementation with wild type TRF2 is similar to that of MEFs expressing the endogenous TRF2 (data not shown; see for instance Figure 5). Importantly, there was no difference in t-loop frequency in cells expressing wild type TRF2 and TRF2ΔTR (Figure 1(c,d)). Similarly, TRF2ΔTR was fully proficient in sustaining normal t-loop levels in Cre-treated TRF1 F/F TRF2 F/F Ku80 -/-MEFs (Supplemental Figure 1(ac)).
Consistent with the ability of TRF2ΔTR to form t-loops, this allele was fully proficient in repressing the formation of telomere fusions when expressed in Lig4-proficient cells (Figure 1(eg)). TRF2ΔTR was also largely proficient in repressing the activation of ATM signaling that occurs in TRF2 F/F MEFs treated with Cre ( Figure 1(h,i)). The residual DDR response at telomeres containing TRF2ΔTR instead of wild type TRF2 is most likely due to the previously noted minor role of TIN2 in repression of ATM [30].
These data suggest that TRF2 is the only component of shelterin necessary and sufficient for t-loop formation, although it does not exclude the involvement of non-shelterin factors. One caveat in our experiments is that the expression level of the exogenous versions of TRF2 is much greater than the endogenous TRF2 (see Supplemental Figure 1(d)), possibly masking a subtle defect in t-loop formation. A second caveat is that cells expressing TRF2ΔTR have telomeres with elongated 3 overhangs [28], although there is no evidence that longer overhangs enhance the detection of t-loops [19].

Two forms of TRF2 from a single mRNA equally protect telomeres
When human TRF2 cDNAs were originally isolated from cDNA libraries, most clones lacked the Basic N-terminus of the ORF, presumably due to the inherent problems in copying this extremely G/C-rich part of the TRF2 mRNA [2,3]. However, extensive screening of a number of cDNA libraries yielded a single TRF2 cDNA with a Kozak consensus ATG start codon [3], which was assumed to represent full-length TRF2. This cDNA and an equivalent cDNA for mouse TRF2 have been used for all in vivo and in vitro studies of TRF2 to date, including complementation of the mouse TRF2 KO phenotype [4].
However, according to the current database annotation (www.Ensembl.org), the human and mouse TERF2 loci express a mRNA that encodes a longer ORF that starts with a Kozak consensus ATG 42 aa of the previously identified ATG (Figure 2(a)). This N-terminal extension is highly conserved (Figure 2(a)). In immunoblots that are well resolved, our TRF2 antibody detects two bands that disappear when Cre is used to delete TRF2, potentially representing the use of both ATGs in the TRF2 mRNA ( Figure 2(b)). We refer to these forms as TRF2 Long and Short. Expression of the original mouse TRF2 cDNA that starts with the second ATG results in a protein that co-migrates with TRF2 Short (Figure 2(c); Supplemental Figure 1d). To determine whether the upper band represents the fulllength TRF2, we generated a construct containing the full N-terminus of TRF2 starting with the first ATG (TRF2 Long). Expression of this construct yielded a single protein that co-migrated with the Long form of TRF2 (Figure 2(c)). This result raised the question of why the second ATG in our construct is not used, whereas presumably the second ATG of the endogenous TRF2 mRNA is frequently used.
The only difference between our full-length TRF2 mRNA and the endogenous mRNA is the 5 UTR. Inspection of the mouse and human 5 UTRs indicates that it can form a stem-loop structure with the region surrounding the first ATG by FISH using the indicated MEFs infected as in (e) and treated with Cre (96 h). (g) Quantification of telomere fusions detected as in (f). Scoring was performed in three independent experiments with 10 metaphase per experiment. Averages, SDs, and significance as in (d). *** p < 0.001. (h) Immunoblot for TRF2 expression in the indicated MEFs infected with vector, TRF2, and TRF2ΔTR before and 96 h after Cre treatment. * nonspecific band. i, TIF assay using 53BP1 IF for detection of the telomere damage response on cells as in (h). (j) Quantification of the TIF response as in (i). Averages and SDs from three independent experiments with 50 nuclei each. Significance as above. (Supplemental Figure 2a,b). Stability prediction algorithms (Sfold and mfold) showed that these sequences adopt stable folds with ΔG values of approximately −30 kcal/mol. Such secondary structure could potentially repress the use of this ATG and lead to the expression of both the Long and the Short form of TRF2 from a single mRNA. The secondary structure in the 5 end of the TRF2 mRNA is highly conserved and in each case the use of the first ATG is likely to be impeded (Supplemental Figure 2a,b). If the stem-loop structure surrounding the first ATG indeed represses initiation of translation, it is predicted that initiation at the second ATG involves an IRES located upstream of the ATG. Further work will be required to identify this IRES.
Given the conserved nature of the 5 end of the TRF2 mRNA and its presumed effect on the expression of the two TRF2 isoforms, it was prudent to test the ability of TRF2 Long to protect telomeres. We expressed both forms (untagged) in MEFs from which the endogenous TRF2 could be deleted with Cre (Figure 2(c)). Both forms of TRF2 were found to be equally proficient in repressing of ATM signaling at telomeres and both repressed the NHEJ of telomeres equally (Figure 2(dg)). Unsurprisingly, the two TRF2 isoforms also showed the same ability to support t-loop formation after deletion of the endogenous TRF2 (Figure 2(h,i)).
The longer isoform of TRF2 contains a conserved SQ site, the target site of the ATM and ATR kinases (Figure 2(a)). We, therefore, tested whether mutation of the Serine residue to alanine or to the phospho-mimetic glutamate affects the function of TRF2 (Figure 2(c)). Neither mutation affected the ability of TRF2 to repress ATM signaling and NHEJ (Figure 2(cg)), although the S23D mutant showed a very minor increase in TIF response. Both mutant versions of TRF2 were also capable of supporting t-loop formation in cells lacking the endogenous TRF2 (Figure 2(h,i)).
We conclude that there is no discernable difference between the ability of the two isoforms of TRF2 to protect telomeres, validating past studies in which only the shorter form of TRF2 was used for complementation and mutational analysis. Since both isoforms function equivalently at telomeres, it is unlikely that the potentially regulatory secondary structure in the TRF2 mRNA is important for telomere protection. However, there is a cytoplasmic splice variant of TRF2 (TRF2-S) that lacks the Myb DNA binding domain and the NLS. TRF2-S regulates the REST transcriptional repression complex in neuronal progenitors [32]. Potentially the two isoforms of TRF2-S that are predicted to result from differential use of the two ATGs behave differently in this regard. If so, this could explain the evolutionary conservation of the secondary structure in the 5 end of the TRF2 mRNAs.

Testing the topological model for TRF2-mediated t-loop formation
To test whether TRF2 forms t-loops by changing DNA topology, we used the Topless equivalent of mouse TRF2 (containing the same 7K2 R to A mutations as in human Topless TRF2 [25]) in a setting where the endogenous TRF2 can be deleted with Cre ( Figure 3). We included a version of the Topless allele that lacks the Rap1 binding site (ToplessΔR) in order to test whether the absence of Rap1 promotes telomere fusions as reported [25]. T-loop analysis showed that both Topless and ToplessΔR were severely defective in forming t-loops, consistent with prior data [25] ( Figure 3(ac)). Both versions of TRF2 were also defective in the repression of ATM signaling at telomeres (Figure 3(d,e)), again consistent with prior data [25] To understand why Topless TRF2 performs poorly in telomere protection, we examined its localization to telomeres more closely. Both Topless and ToplessΔR can be readily detected at telomeres in paraformaldehyde fixed samples (Figure 4(a,b)). However, when cells were preextracted with Triton-X100 and fixed with MeOH, these alleles show a striking deficiency in telomeric localization (Figure 4(a,b)). Quantification of the data indicates a significant reduction in the ability of Topless TRF2 to withstand pre-extraction compared to wild type TRF2, suggesting a diminished binding capacity (Figure 4(c)). ChIP analysis of formaldehyde cross-linked chromatin also showed diminished presence of the Topless alleles in association with telomeric DNA. In part, the reduced telomeric localization may be due to the lower expression level of the Topless alleles. In addition, it is not excluded that the Topless mutants have diminished or altered DNA binding activity. Although gel-shift experiments with human Topless TRF2 showed roughly the same affinity for a short telomeric substrate as the wild type TRF2, some of the R and K mutations in the TRFH domain altered the affinity [25]. Additional tests will be required to determine whether Topless TRF2 engages telomeric DNA in an altered manner.

T-loop frequencies are similar in G1, S, and G2
To gain further insight into the mechanism of t-loop formation, we asked whether this process is cell cycle regulated. The frequency of t-loops was examined in wild type MEFs that were treated with various agents to enforce enrichment in G1 (Lovastatin, Figure 5 (ad)), S phase (Lovastatin followed by release into S phase, Figure 5(eg)), or G2 (RO-3306 and Nocodazole, Figure 5(il)). In each experimental setting, the cell cycle profile of the resulting cell populations was determined by FACS analysis of BrdU labeled cells. Furthermore, for each setting, we determined by immunoblotting that the expression level of TRF2 was not altered ( Figure 5). Measurements of t-loop frequencies indicated the lack of a significant difference when comparing asynchronous populations to cells enriched in G1, S, or G2. We have not examined t-loop frequencies in mitosis but prior work indicates that they exist in that cell cycle stage as well [33].
The lack of cell cycle regulation of the t-loop structure precludes models for t-loop formation that invoke factors specifically expressed in S phase (e.g. certain components of the homologous recombination pathway). However, our data is not in disagreement with the idea that t-loops are resolved during the replication of telomeric DNA [34,35], since opening of t-loops would represent a short-lived state that takes place at each telomere at a different time during S phase.
The results are also informative with regard to the question of why t-loops are never observed at 100% of the telomeres analyzed. One explanation for the finding that t-loops usually are detected at less than half of the scored molecules might have been that t-loops are not present in S phase. Our data argue against this possibility although it is likely that individual t-loops are opened briefly in S phase to allow replication to proceed [34]. We consider it more likely that the t-loop frequencies are an underestimate due to breakage of telomeric DNA during the spreading technique, resulting in one linear and one t-loop molecule. Two other sources of underestimated t-loop frequencies are insufficient cross-linking at the base of the t-loop and the inability to score t-loops if the loop part is small [19].

Rad51 is not required for t-loop formation
We considered that Rad51 could facilitate t-loop formation by TRF2 since this recombinase is predicted to enable the strand-invasion of the 3 overhang. Indeed, Rad51 has been implicated in this role based on in vitro experiments [26]. To test the role of Rad51 we used CRISPR/Cas9 gene editing in bulk cell populations. The effective deletion of Rad51 was evaluated through immunoblotting and based on the absence of Rad51 foci in IR treated cells (Figure 6(ac)). The removal of Rad51 led to slightly elevated DNA damage response at telomeres (Figure 6(d)) and a minor increase in telomere fusions (Figure 6(e)). Both phenotypes may be due to a role for Rad51 at sites of replication stress since telomeres often experience replication problems. Nonetheless, the phenotypes of Rad51 deletion did not resemble the effects of TRF2 deletion and commensurate t-loop loss. Consistent with these findings Rad51 deletion did not alter the frequency of t-loops in the cells (Figure 6(h,i)). Although these experiments do not exclude that Rad51 contributes to t-loop formation in a redundant fashion, the data indicate that Rad51 is not required for t-loop formation.

Conclusions
Here we show that t-loop formation requires only a single component of shelterin, TRF2. This result indicates that t-loop formation is dependent on a feature of TRF2 itself or on a second factor that is brought to telomeres by TRF2. We tested one feature of TRF2, its ability to wrap telomeric DNA, but were unable to verify that this attribute is relevant to t-loop NUCLEUS formation due to the hypomorphic nature of the critical Topless mutant of TRF2. To better test the topology model for t-loop formation, Topless versions are needed that show normal localization to telomeres and retain the ability to repress NHEJ. We also considered the possibility that t-loop formation is not due   to an intrinsic feature of TRF2 but depends on a TRF2-interacting partner. We excluded the most obvious candidate, Rad51, as being involved in t-loop formation. The extensive proteomics efforts on TRF2 complexes and telomeric chromatin have not nominated obvious additional candidates (e.g. [36][37][38][39][40]. Therefore, the mystery of TRF2-mediated t-loop formation remains unsolved and other models should be pursued.

Altered forms of mouse TRF2
The version of TRF2 lacking the Rap1 and TIN2 bindings sites (ΔTR, Δ350-365 [41] and Δ284-297 [29]) was generated by PCR-based mutagenesis on a version of TRF2 lacking the Basic domain and then cloned into a pLPC vector containing the Basic domain with an N-terminal Myc tag and C-terminal BamHI site. For generating the long version of TRF2, the whole N-terminal sequence (including the Basic domain) was synthesized by Genewiz and cut out of a plasmid with BglII and BamHI. This fragment was ligated to the BamHI site of a pLPC plasmid containing the rest of TRF2 with a BamHI site at the N-terminus of the TRFH domain [31]. Although both ATGs are present in this construct, only the first ATG appears to be used in vivo. The TRF2 Topless mutant was generated by site-directed mutagenesis with the following aa changed to alanine: R69, R99, K158, K173, K176, K179, K241, K242, and K245. Mutagenesis was performed on mTRF2 lacking the Basic domain in pBluescript. The mutant version was then moved into a pLPC vector containing the Basic domain with an N-terminal Myc tag and a C-terminal BamHI site. This Topless version was used to delete the Rap1 binding site (Δ284-297) by PCR-based mutagenesis.

Cell cycle experiments
MEFs were treated with either 9 µM RO-3306 (Sigma, SML0569-5 M) for 12 h and released into 1 µg/ml Nocadozole (Sigma, M1404) for 2 h, or with 40 µM Lovastatin (Apex Bio A4365) for 36 h, or with 40 µM Lovastatin for 36 h and released into 400 µM Mevalonic Acid (Sigma, 90469-10 MG) for 16 h, or with DMSO as a control. Flow cytometry was conducted on an Accuri C6 and cells were prepared as follows: 1 h prior to harvest cells were treated with 10 µM BrdU and 500,000 cells were then harvested and fixed overnight in −20°C 70% EtOH. Cells were three independent experiments. Significance as in Figure 1. (f) Examples of metaphases from cells treated with an sgRNA to Luciferase or Rad51. (g) Quantification of telomere fusions in MEFs treated with sgRNAs to Luciferase or Rad51. Data show averages and SDs from 3 experiments. Significance as in Figure 1. (h) Examples of t-loops detected in cells treated with the indicated sgRNAs as in (b). Scale bars: 2.5 μm. (i) Quantification of t-loop frequencies. Averages and SDs from three experiments with 100 molecules scored for each. Lack of significance is indicated (determined using a two-tailed unpaired t-test). denatured and permeabilized with 2 N HCl/0.5% Triton X-100 for 30 minutes, then neutralized with 0.1 M Sodium Borate. Cells were blocked with 0.5% BSA/0.5% Tween-20 in PBS and incubated with a FITC-conjugated anti-BrdU antibody (BD Biosciences #347583), then resuspended in 2 mM EDTA, 0.5 μg/ml RNAse A, and 5 µg/ml Propidium Iodine for FACS analysis. The data were processed with Flojo software.