To address the specificity of the increased X-structures at telomeres, X-structure levels were examined in the ribosomal DNA (rDNA). Although rDNA X-structures were elevated by sgs1 mutation, consistent with elevated rDNA recombination and higher X-structure levels in sgs1 mutants in earlier reports [29,30], telomere shortening caused by tlc1 mutation caused no increase in levels (Figure 2), indicating specific accumulation at telomeres with senescence. X-structures at several genomic loci have been found to accumulate in late S or G2/M phases of the cell cycle. Because senescent telomerase mutants similarly accumulate in the G2/M phase of the cell cycle [15,31,32], this raised the possibility that elevated telomere X-structures in senescent cells might simply reflect a cell cycle effect. However, the lack of increased rDNA X-structures with senescence indicates that this is not the case. Similarly, the lack of increased telomere X-structures in sgs1 mutants indicates that perturbations in cell cycle progression reported by some investigators in these mutants [29,33–35] do not explain the increases that occur during senescence. Rather, elevated telomere X-structures are somehow related specifically to telomere shortening, and Sgs1p plays a role in attenuating the accumulation of telomere X-structures in this context. Such a role of Sgs1p might be to prevent the formation of telomere recombination intermediates or to facilitate their resolution.

We considered several possible identities for the X-structures, including Holliday junctions (HJs; and structurally similar regressed replication fork or “chicken-foot” structures), convergent replication forks, hemicatenanes (HC), and so-called rec-X structures (Figure 3A). HC have an interlink between single strands from two duplexes, and such a link forms in a RAD52-independent fashion between sister chromatids behind advancing replication forks, whereas rec-Xs are thought to form in a RAD52-dependent fashion at stalled forks when one nascent strand switches its template transiently for the other nascent strand and then returns to the original template (see also Figure 7A) [27,36–38]. The different types of X-structures can be distinguished based on their susceptibility to cleavage by HJ resolvases such as RuvC, their ability to branch migrate, and their dependence on RAD52. To characterize the X-structures, we first tested whether they could be cleaved by the HJ-specific resolvase RuvC [39]. RuvC (350 ng) did not cleave telomere X-structures from tlc1 sgs1 cells, despite fully cleaving a synthetic X-junction added to the same reactions containing yeast DNA (Figure 3B, top left and right, and bottom right panels, and unpublished data). Five-fold higher levels of RuvC (1.75 μg) partially degraded both the replication arc and X-structures, but had no preferential effect on the X-structures (Figure 3B, bottom left panel). Therefore the X-structures are apparently not HJs. Incubation of gel slices from the first dimension in branch migration buffer at 65 °C prior to running the second dimension caused selective loss of the spike and retention of the replication arc in tlc1 and tlc1 sgs1 mutants (Figure 3C). The X-structures therefore cannot be convergent forks, because replication forks do not branch migrate [25], thus leaving HC or rec-Xs as possibilities because they do branch migrate [27,36]. Further, the X-structures branch migrated in the presence of Mg⁺², consistent with a HC or rec-X, but not HJ, identity (unpublished data) [27].

To test RAD52 dependence of the X-structures, we first re-examined the effect of rad52 mutation on senescence of tlc1 and tlc1 sgs1 mutants. This was done because our previous studies used a rad52 disruption allele beginning after amino acid 167 [22,40], potentially allowing for expression of the N-terminus. Because N-terminal fragments of Rad52p have been shown to be sufficient for DNA binding and strand-annealing activities in vitro [41,42], this raised the possibility that the disruption allele might have residual activity. However, a full-deletion rad52 allele yielded the same results as the disruption allele: rad52 deletion sped senescence of tlc1 mutants, and additional sgs1 mutation had no further effect on the rate of senescence (Figure 4A), confirming that RAD52 is epistatic to SGS1 during senescence. As reported previously [21,43], RAD52 was required for survivor formation in tlc1 mutants, because they depend on recombination. Curiously, at later times than those shown, tlc1 sgs1 Δrad52 cells were actually able to form poorly growing survivors, as will be described elsewhere (J. Y. Lee and F. B. Johnson, unpublished data). Comparison on 2DGE of PD-matched cultures of tlc1 sgs1 with tlc1 sgs1 Δrad52 cells showed a significant 47% reduction, but not elimination of telomere X-structures caused by the rad52 deletion (Figure 4B and unpublished data). RAD52 deletion also had no effect on X-structures in TLC1+ and early generation tlc1 mutants prior to significant rise in X-structure levels (Supplementary Figure 2 in Protocol S1, and unpublished data). We conclude that the elevated level, but not the basal level, of X-structures in tlc1 sgs1 mutants is dependent on RAD52. Rec-X formation is RAD52-dependent, whereas HC formation is not [27,36], and so, as detailed in the Discussion, a reasonable interpretation of the data is that the basal X-structures are HC and the elevated levels represent rec-Xs formed secondary to stalled replication at telomeres.

To further test the idea that the elevated level of X structures might correspond to rec-Xs, we examined the dependence of the telomere X-structures on RAD53. At stalled replication forks, Rad53p functions to stabilize the replisome, signal the intra-S phase checkpoint, and facilitate the resumption of replication [6,44]. The kinase-defective rad53K227A allele has been shown to lead to the loss of rec-X structures at stalled forks, including suppression of elevated rec-X levels in sgs1 mutants treated with methane methyl sulfonate [27,36]. tlc1 rad53K227A mutants senesced faster than tlc1 mutants (Figure 5A), consistent with a role for Rad53p in preventing replication fork collapse at telomeres and thus premature senescence in the absence of telomerase. Further, tlc1 sgs1 rad53K227A mutants did not senescence significantly faster than tlc1 sgs1 mutants, as expected if Sgs1p and Rad53p function in the same pathway to prevent rapid senescence. tlc1 sgs1 rad53K227A mutants showed a significant 2-fold reduction in X-structure levels compared with PD-matched tlc1 sgs1 mutants (Figure 5B and ​and5C),5C), supporting the hypothesis that rec-Xs account for the elevated X-structure levels.

The greater accumulation of X-structures in tlc1 sgs1 mutants could reflect either increased formation or reduced resolution of recombination intermediates in the absence of Sgs1p. To address these alternatives, and as a second test for telomere recombination during senescence, we used telomere PCR [45] to clone and sequence examples of telomeres from a single chromosome end in the populations of senescing cells. The yeast telomere repeat is imperfect, with an approximate consensus of (TG_1–3)_n, and individual telomeres thus differ in their precise sequence. Telomerase adds new and variable sequences to telomere ends, but in tlc1 mutants, the sequence is fixed and typically does not change as a telomere shortens. However, occasional recombination events append new sequences at telomere ends during senescence [46]. If Sgs1p resolves telomere recombination intermediates and thus allows continued cell division, telomere recombinants should be reduced in tlc1 sgs1 mutants (because these cells would arrest and yield no progeny), whereas if it inhibits the formation of telomere recombination intermediates, an increase in recombinants would be predicted (because recombination events would be more frequent and recombinant cells would yield progeny). Genomic DNA was isolated from tlc1 and tlc1 sgs1 mutants at 44 PD after loss of telomerase, far in advance of survivor formation. DNA ends were polyC tailed, and then amplified by PCR using a polyG primer specific to the tail and one specific to sequences internal to the telomere on the left arm of Chromosome I (telomere I-L). Sequence analysis of 434 tlc1 and 439 tlc1 sgs1 cloned products showed that, as expected, the majority of the G-rich strands from telomere I-L had a distribution of lengths, which ranged from 8–178 nucleotides (nt), but were identical in sequence over their shared lengths (Figure 6A). The different lengths in the population of senescing cells reflect the stochastic natures of senescence and telomere shortening, as observed previously [46]. Analysis of telomere sequences that diverged from the original showed that 5.1% (22/434) of sequences in tlc1 mutants were recombinant, similar to the 6.6% reported previously [46]. There was a trend toward fewer recombinants in tlc1 sgs1 mutants, with recombinants accounting for only 3.9% (17/439) of telomeres (p = 0.196). Recombinants included apparent cases of inter-telomeric recombination as well as intra-telomere recombination resulting in duplication of telomere sequences (Figure 6B). Importantly, there was a significant difference in the distribution of recombinants as a function of the length of the nonrecombined telomere repeat tract, with recombinant telomeres occurring at a significantly reduced frequency at longer telomeres in the tlc1 sgs1 mutants (Figure 6A). Comparing telomeres with nonrecombinant tract lengths of greater than 85 nt, there was a 2.6-fold reduction in the frequency of recombinant telomeres in the double mutants (14/283 vs. 5/264; p = 0.026). As detailed in the Discussion, these results can be explained if Sgs1p facilitates the resolution of telomere recombination intermediates and if unresolved intermediates in tlc1 sgs1 mutants lead to cell cycle arrest, thus inhibiting the accumulation of progeny bearing recombinant telomeres.

We observed an accumulation of X-shaped structures at telomeres in tlc1 sgs1 mutants. Because RuvC cannot selectively cleave these X-structures, they do not appear to be HJs. This is supported by our recent report that the C-terminal 200 amino acids of Sgs1p are dispensable for slowing senescence [22]; this C-terminus contains the HRDC domain, which is important for HJ binding and double-HJ dissolution [47–49], arguing that HJ-targeted functions of Sgs1p are not involved in slowing senescence. These X-structures can branch migrate, indicating that they cannot be convergent replication forks, but this is consistent with them being HC or rec-X structures. In S. cerevisiae, HC have been observed to form behind replication forks in a RAD52-independent fashion [36,38]. Liberi et al. [27] suggested that at a stalled fork, resumption of replication might use a HC to facilitate template switching whereby one nascent strand leaves the parental template and copies the other nascent strand (to bypass the cause of the stall) before returning to the original template in a RAD52-dependent step, thus forming a rec-X structure. Our finding that the elevated level, but not the basal level, of X-structures in tlc1 sgs1 mutants is RAD52- and RAD53-dependent is consistent with the elevated level structures being rec-Xs and the basal level structures being HC. According to this model (Figure 7A), the stalling of replication forks that occurs naturally in the telomere repeats [24,50] might be somehow enhanced by changes related to telomere shortening (see below) and thus might lead to rec-X formation and eventual resolution by Sgs1p; in the absence of Sgs1p, rec-X structures would accumulate. This is analogous to the recently reported accumulation of rec-X structures at non-telomeric loci in sgs1 mutants after the stalling of replication forks by methyl methane sulfonate [27]. Sgs1p would be expected to function in tandem with its Top3p cofactor to effect strand transfer reactions that would enable resolution of the rec-X structure [27,51,52], consistent with our finding that such cooperation is required to prevent rapid senescence [22]. If unresolved, rec-Xs might lead directly to cell cycle arrest; if resolved by other means (e.g., nucleases), the shortened or aberrantly structured telomere ends might hasten the onset of senescence. In TLC1+ cells, telomerase could repair such ends, thus explaining the normal telomere length in sgs1 mutants and the synergy of sgs1 mutation with tlc1 mutation to accelerate senescence. Previously, we observed that senescent tlc1 sgs1 mutants appear unable to segregate nuclei between mother and daughter cells [15], and a possible explanation is that unresolved recombination intermediates interfere with chromosome segregation. We note also that suppression by Sgs1p of X-structure accumulation at stalled replication forks was recently shown to cooperate with a parallel pathway that is dependent on SUMOylation [18]. This might explain our recent observation that, like Sgs1p, Slx5p and Slx8p are required to prevent rapid senescence of tlc1 mutants, because Slx5p and Slx8p function in parallel with Sgs1p for cell viability and also show genetic interaction with SUMO pathway factors [22,53].

It is not yet clear why telomere shortening should increase X-structure levels at telomeres, although one possibility is that changes in chromatin might contribute to increased replication fork stalling and thus rec-X formation. For example, decreased Rap1p binding at shortened telomeres might allow telomere repeats to adopt DNA structures involving hydrogen bonding between guanines, e.g., G-quadruplexes, that could impede replication. The recent demonstration of impaired telomere replication in S. pombe cells lacking the Taz1 telomere repeat binding protein supports this idea, and led the authors to also propose that Rap1p might serve a similar function to facilitate telomere replication in S. cerevisiae [54]. Of note, however, a mutant Rap1p protein lacking the C-terminus does not impact telomere fork stalling [24], although this mutant possesses the N-terminal DNA binding domain and so might have retained the function proposed to facilitate replication. We further note that we did not observe in senescing tlc1 or tlc1 sgs1 cells an increase in apparent stalling near the 2N spot, corresponding to the telomere repeats, although this might be explained by efficient conversion of stalled forks into rec-X structures. A second possibility is that a shortened and less heterochromatic telomere might be more accessible to recombination factors like Rad52p, thus facilitating rec-X formation. The elevated recombination rates at shortened telomeres in K. lactis and telomerase knockout mice are consistent with these possibilities [55,56].

During senescence, Sgs1p may inhibit the formation of telomere recombination intermediates, or facilitate their resolution. Using telomere PCR and sequencing, we observed a significantly decreased frequency of recombinants occurring at longer telomeres in tlc1 sgs1 mutants, supporting the model that Sgs1p helps resolve telomere recombination intermediates into mature products. For this reason, we suggest that decreased resolution, rather than increased formation of X-structures explains their higher levels in tlc1 sgs1 mutants. Two aspects of the PCR assay used to detect the recombinant telomeres must be understood to explain why recombinants are less frequent at longer telomeres in the absence of Sgs1p. First, the number of cell divisions after loss of telomerase at which a recombinant arises will influence the apparent frequency of that event: productive recombination events that occur early and give rise to progeny will remain at a frequency approximately equal to that at the time of their occurrence, whereas events that occur late will appear at a lower frequency that reflects the larger size of the pool of cells at that later time point (Figure 7B, left vs. right). Although only one time point was examined for each strain, the stochastic natures of senescence and telomere shortening caused some cells to be closer to senescence than others, and so information spanning a large range of telomere lengths was obtained. The second aspect of the telomere PCR assay that must be appreciated is that it will detect both resolved products and unresolved intermediates, and therefore, any stalled recombination intermediates in cells lacking Sgs1p will still be observed; this fact minimizes the measured difference in recombinants between tlc1 and tlc1 sgs1 mutants when telomeres of all lengths are examined (because recombinants forming at short telomeres are expected to suffer little from the sgs1 defect; see below). Telomere recombination appears to increase as telomeres shorten (Figure 1D and [28]), yet the frequency of recombinants measured by telomere PCR in tlc1 cells was not greater at shorter than at longer telomeres. A reasonable explanation is that the recombination events at short telomeres were more likely to have occurred in cells that were closer to senescence and so gave rise to fewer progeny than the cells experiencing recombination at long telomeres. These competing effects of more frequent recombination at short telomeres but fewer recombinant progeny arising from cells with short telomeres could balance each other in the tlc1 mutants so that the distribution of recombinants is similar among telomeres of all sizes. In contrast, if Sgs1p is required for efficient resolution of recombination intermediates and if unresolved intermediates cause cell cycle arrest, then cells with long telomeres, and thus high replicative potential, would be most affected by stalled recombination events; tlc1 sgs1 mutants arrested by stalled recombination intermediates at long telomeres will become diluted by the other dividing cells (Figure 7B, middle). Cells with short telomeres are unlikely to divide much further regardless of the outcome of a telomere recombination event, and thus absence of Sgs1p would have relatively little effect on the measured frequency of recombinants at short telomeres (Figure 7B, right). This explains why the decrease in recombinants in the tlc1 sgs1 mutants occurs preferentially at longer telomeres. As an interesting aside, by this view, the distribution of recombinants among telomeres of different lengths in tlc1 sgs1 mutants most accurately reflects the propensity of short telomeres to engage in recombination because this distribution is not skewed, as it is in tlc1 mutants, by the opposing effect of recombination at longer telomeres tending to occur in cells with greater remaining replicative potential and thus giving rise to more progeny. Consistent with the interpretation that stalled recombination events in tlc1 sgs1 mutants lead to permanent cell cycle arrest, WRN is required in cultured human cells for the resolution of recombination intermediates that enable cells to generate viable progeny [57].

If tlc1 mutants can complete telomere recombination and give rise to viable progeny, then repeat examples of the same recombination event should be detectable. Indeed, four independent examples were obtained in the tlc1 cells (indicated by the number sign [#], Figure 6A). No such repeat events were observed in the tlc1 sgs1 cells, consistent with telomere recombination often being a terminal event in the absence of Sgs1p. Furthermore, no such events were observed in the shortest (<85 nt) telomeres of tlc1 mutants, consistent with the recombination events at short telomeres occurring in cells that are near the end of their lifespan.

The action of Sgs1p during senescence need not reflect any telomere-specific function, but rather may be one manifestation of a general role in the restart of replication forks stalled for various reasons, for example, hydroxyurea treatment, DNA alkylation by methane methyl sulfonate, or as proposed here, chromatin changes at shortened telomeres. We note, however, that stalled forks in the terminal telomere repeats would be particularly problematic because there is no replication origin distal to the stall to generate a rescuing fork, thus perhaps contributing to the dependence of telomeres on Sgs1p-dependent restart during senescence. Sgs1p helps activate the checkpoint response to DNA damage in S phase, and also helps to stabilize DNA polymerases alpha and epsilon at stalled forks [35,58]. The former function, but not the latter, is thought to occur in collaboration with Rad53p [58]. Nonetheless, Rad53p appears to help stabilize stalled replication forks [27,36,59,60], although the extent to which this reflects stabilization of DNA polymerases [35,61], the MCM helicase [62], or other functions, and the degree to which Sgs1p is required for these functions, are not resolved at present. The reduction in X-structure levels caused by the rad53K227A allele and modest acceleration of senescence in tlc1 rad53K227A mutants is consistent with the model that rec-X–dependent fork restart contributes to optimal telomere replication during senescence (Figure 7A). The larger effect of sgs1 mutation on senescence may reflect a hypomorphic effect of the rad53K277A allele with respect to HC- and rec-X–mediated fork rescue, such that some stalled forks are still routed through this pathway and thus depend on Sgs1p function. Alternatively, the capacity of Sgs1p to stabilize stalled forks may be greater than that of Rad53p. Our findings leave open the possibility that replisome stabilization by Sgs1p may contribute to slowing senescence, in addition to the proposed role in rec-X resolution.