Molecular architecture of G-quadruplex structures generated on duplex Rif1-binding sequences

G-quadruplexes (G4s) are four-stranded DNA structures comprising stacks of four guanines, are prevalent in genomes, and have diverse biological functions in various chromosomal structures. A conserved protein, Rap1-interacting factor 1 (Rif1) from fission yeast (Schizosaccharomyces pombe), binds to Rif1-binding sequence (Rif1BS) and regulates DNA replication timing. Rif1BS is characterized by the presence of multiple G-tracts, often on both strands, and their unusual spacing. Although previous studies have suggested generation of G4-like structures on duplex Rif1BS, its precise molecular architecture remains unknown. Using gel-shift DNA binding assays and DNA footprinting with various nuclease probes, we show here that both of the Rif1BS strands adopt specific higher-order structures upon heat denaturation. We observed that the structure generated on the G-strand is consistent with a G4 having unusually long loop segments and that the structure on the complementary C-strand does not have an intercalated motif (i-motif). Instead, we found that the formation of the C-strand structure depends on the G4 formation on the G-strand. Thus, the higher-order structure generated at Rif1BS involved both DNA strands, and in some cases, G4s may form on both of these strands. The presence of multiple G-tracts permitted the formation of alternative structures when some G-tracts were mutated or disrupted by deazaguanine replacement, indicating the robust nature of DNA higher-order structures generated at Rif1BS. Our results provide general insights into DNA structures generated at G4-forming sequences on duplex DNA.

G4 structures formed on ssDNA have been extensively studied, and diverse structures and topologies have been described at the atomic level (24). However, those generated on duplex DNA have been poorly understood despite the highly prevalent presence of potential G4 structures on duplex genomes.
Recently, we have reported that G4 may play a major role in regulation of replication timing through its recognition by a conserved nuclear protein, Rif1 (25). Rif1 is a HEAT repeatcontaining protein, which was originally discovered as a telomere-binding factor in yeasts (26,27). Rif1 was shown to be required for temporal and spatial regulation of DNA replication in both yeast and human cells (25, 28 -32). Rif1 affected chromatin loop sizes and, more recently, was shown to affect the chromatin interactions within the domains that define the replication timing (33). We have identified Rif1-binding sites (Rif1BS) along the genome of fission yeast and have discovered the presence of a conserved sequence motif (Rif1 consensus sequence (Rif1CS)) in Rif1BS (25). Single-stranded Rif1BS sequences are capable of forming G4 in vitro in a manner dependent on guanine sequences in Rif1CS. Rif1CS, containing 5-6 runs of guanines, frequently appear as a head-to-tail dimer, in which the two Rif1CS are separated by a ϳ100-bp segment, and both Rif1CS are required for efficient chromatin binding of Rif1. Thus, a DNA segment of over 100 bp is involved in binding of Rif1 to chromatin.
There are additional runs of guanines in Rif1BS, and these guanine residues have been shown to contribute to the formation of G4 structures on ssDNA and also to in vivo chromatin binding of Rif1 (25). However, the sequences of Rif1 BS are unorthodox in that G-tracts are scattered over a 100-bp segment in addition to the presence of G5 or G6 tracts in Rif1CS, whereas only a 15-25-nucleotide ssDNA is sufficient to form a G4 in a test tube (24). As a consequence, Rif1-binding sequences are significantly diverged from the known consensus sequence for G4 (G Ն3 N x G Ն3 N x G Ն3 N x G Ն3 , where x ϭ 1-7). In addition, two copies of Rif1CS are sometimes positioned in a head-to-head orientation, causing long G-tracts to be present on both strands.
The intercalated motif (i-motif) structure is formed through a stack of intercalating hemiprotonated C-neutral C bp (Cϩ:C) under slightly acidic pH conditions (34). A recent report (35) indicates the in vivo presence of i-motif through the use of a specific antibody. Naturally, the complementary strand of G-rich sequences capable of forming G4 has a capacity to generate i-motif, although it has not been clear whether these two structures coexist on a duplex DNA.
Because G4 is generated on duplex DNA in vivo, it would be important to determine the structures of G4 generated on dsDNA in competition with more stable B-type helical structure. The results are consistent with the presence of G4-like structures on the G-rich strand of the duplex Rif1BS upon heat denaturation/reannealing. The nuclease sensitivity assays demonstrate the existence of unexpectedly long loop segments in the generated G4-like structure. They also indicate the generation of a higher-order structure on the other C-rich strand, which strictly depends on the G-tracts on the G strand and is not likely to be i-motif. We also show that Rif1BS have a potential to form multiple G4 structures on the G-rich strand, and in some cases G4 could be formed also on the C-strand. We suggest that Rif1BS may generate G4 structures that may undergo dynamic transition from one structure to another through its potential to generate multiple G4-like structures on one or both strands.

ssDNAs from Rif1BS containing a single Rif1CS are able to form G4-like structures
Rif1BS contains at least two Rif1CS, with head-to-tail orientation in 75% of the cases, and the two Rif1CS are separated by a ϳ100-bp segment on average. In addition to the 5-6 runs of guanines on Rif1CS, there are often additional runs of guanines (G) in the Rif1BS on the same strand. Therefore, we have made ssDNAs containing various sequences from Rif1BS I:2663 and Rif1BS II:4255 , two strong Rif1BS, and analyzed them for heatinduced formation of higher-order structures and binding of Rif1 (Fig. 1). Most of the sequences containing the G-tracts of Rif1CS were capable of forming slowly migrating forms on

G-quadruplex structures on a duplex DNA
polyacrylamide gels containing KCl and PEG 200, a reagent that facilitates formation of G4 with a molecular crowding effect (38).
Rif1 bound to almost all of these structures. Slow migration on PAGE suggests that these ssDNAs form higher-order structures that probably involve multiple molecules of the oligonucleotide DNA used. It should be noted that those generating slowly migrating forms always contain G-tracts of Rif1CS. Indeed, Rif1-1 that does not contain a long G-tract but contains two 3Gs did not generate slowly migrating forms and was not bound by Rif1 (Fig. 1, lanes 1 and 2). We have analyzed the structures of Rif1-1 to Rif1-10 by CD (Fig. S1). G4 DNA can adopt various topologies that are known to exist for G4 DNA. Those include parallel, anti-parallel, hybrid, and mix (24). Different topologies can be identified by measuring CD spectra. In the case of a typical anti-parallel structure, positive cotton effects are observed at 290 and 240 nm, and a negative one is observed at 265 nm. The results indicate that the Rif1BS-derived sequences generate parallel-type G4 structures. This is consistent with our recent finding that Rif1 prefers parallel-type topology over anti-parallel or hybrid topology. 3 These results suggest that sequences from Rif1BS have the potential to form diverse G4-like structures, and a long G-tract in Rif1CS may play a crucial role in generating the higher-order structures that are recognized by Rif1.

Higher-order structures generated on a ssDNA derived from Rif1BS II:4255 : Potential oligomerization of G4
To analyze in more detail the structure generated on these sequences, we took a ssDNA, Rif1-11(wt), derived from Rif1BS II:4255 , and probed its structures by various nucleases and chemical reagents.
The 81-nucleotide-long Rif1-11(wt) contains two Rif1CS, and derivatives (mut1, mut2, and mut3) carrying mutations at each (mut1 or mut2) or both (mut3) of the Rif1CS were also generated ( Fig. 2A). These ssDNAs generated slowly migrating forms on polyacrylamide gel to different extents. The Rif1-11(wt) generated two major slowly migrating forms (bands b and c) along with a small amount of high-molecular weight aggregates at the top of the gel and a fast-migrating form (band a) that may be a compact G4-like structure composed of a single molecule and is not generated in mut2 or mut3 (Fig. 2B, lane  1). Bands b and c could represent oligomers of G4 structures generated on this sequence. Band c was significantly reduced in Rif1-11(mut1) (Fig. 2B, lane 5), indicating that Rif1CS II-2 is required to generate larger oligomers of G4. Both bands b and c were largely lost with significant increase of large aggregates in Rif1-11(mut2) (Fig. 2B, lane 9), indicating that Rif1CS II-1 is essential for generating G4 that can form oligomers. All of the slowly migrating forms were largely lost in Rif1-11(mut3) except for weak multiple bands (Fig. 2B, lane 13). Rif1 efficiently bound to band c and less efficiently bound to band b and did not bind to band a (Fig. 2B, lanes 2-4), consistent with the idea that Rif1 has a higher affinity to oligomerized G4. Rif1 also Figure 2. G-tracts in Rif1CS affect the structures generated on a ssDNA from Rif1BS II:4255 upon heat denaturation: specific binding of Rif1 and G4-specific antibody. A, 81-nucleotide ssDNAs (Rif1-11) derived from Rif1BS II:4255 . The G-tracts in the two Rif1CS and other 3Gs are shown in red and pink, respectively, and the mutated residues are blue. B and C, gel shift assays with Rif1 (B) or BG4 antibody (C) protein. The ssDNAs shown in A (0.25 pmol), heat-denatured and reannealed in the presence of 50 mM KCl and 40% PEG 200, were used for gel shift assays. Proteins added were as follows. Lanes 1,5,9,and 13, no protein; lanes 2, 6, 10, and 14, 3.5 fmol of Rif1 or 1.4 pmol of BG4; lanes 3, 7, 11, and 15; 7 fmol of Rif1 or 3.5 pmol of BG4; and lanes 4, 8, 12, and 16, 17.5 fmol of Rif1 or 7 pmol of BG4. Complexes were analyzed on 8% PAGE containing 1ϫ TBE, 50 mM KCl, and 40% PEG 200. Bands a, b, and c are estimated to represent, respectively, a monomer, dimer and tetramer of a G4 structure on the basis of the size marker.

G-quadruplex structures on a duplex DNA
bound to the faint slowly migrating forms generated on Rif1-11(mut2) and Rif1-11(mut3) ( Fig. 2A, lanes 10 -12 and 14 -16), albeit with less efficacy. These results suggest that Rif1-11 sequence can form a monomer and two major oligomers of G4, and Rif1CS II-2 is required for formation of the G4 structure that can oligomerize into a larger form. Rif1CS II-1 plays a major role in generating more stable G4 that can oligomerize. The nature of the large aggregates at the well and why they are increased in mut2 is not clear. The structure generated in the absence of Rif1CS II-1 may have a tendency to form a large aggregate.
Evidence showing that these forms are G4 or its related structures was provided by gel-shift assays using BG4, a single polypeptide G4 -specific antibody (39) (Fig. 2C, lanes 2-4). BG4 bound efficiently to Rif1-11(wt), most notably to the band c. BG4 bound to mut1 and mut2, albeit less efficiently. These results show that the Rif1-11 sequence forms a G4 structure that can be efficiently recognized by both Rif1 and BG4, whereas other mutant forms also generate alternative G4 structures, which can be recognized by BG4 and Rif1 to a limited extent.
We then analyzed the structures of Rif1-11 and its mutant derivatives using nuclease probes (S1 nuclease and T7 endonuclease) and chemical modification (DMS modification on guanine residues). The results indicate the formation of higherorder DNA structures expected for G4 in the wt and mutants. Strikingly, alternative G4 structures are formed on mutant derivatives (mut1 and mut2) where one of the two Rif1CS were mutated. A distal mutation in Rif1CS II-2 in mut1 affects the structures involving Rif1CS II-1 (Fig. S2A), which in turn affects the extent of the oligomerization (loss of band c in Fig. 2). The mutation in Rif1CS II-1 in mut2 has a large effect on overall structure, as indicated by dramatic change of S1 and T7 endonuclease sensitivity ( Fig. 2A and Figs. S2 (A and B) and S3), which leads to loss of band b and c. The DMS footprint indicates that an alternate G4 structure may be formed on mut2 by utilizing Rif1CS II-2 (Fig. S2, A and B). Mutation of both Rif1CS II-1 and Rif1CS II-2 (mut3) resulted in almost complete loss of stable higher-order structure (Fig. S2, A and B). These results indicate the critical role of the two long G-tracts in Rif1CS II-1 and Rif1CS II-2 in generating the oligomerized G4 structure that is a preferred target for Rif1. They also indicate the potential of Rif1-11 sequences to adopt alternate G4 structures in response to G-tract mutations (see supporting information for data and a detailed explanation).

Formation of G4-related DNA structures with unusual long loops at Rif1BS duplex DNA after heat denaturation/reannealing
Next, we examined the DNA structures generated on duplex DNA containing Rif1BS. A 447-bp DNA fragment containing sequences from Rif1BS I:2663 was end-labeled and heat-denatured and treated with S1 nuclease, which cleaves ssDNA. The nontreated duplex DNA was largely resistant to S1 nuclease. On the other hand, on the heat-treated DNA, we detected clusters of S1 nuclease-sensitive segments between the Rif1CS I-1 and the 3G-tracts (Fig. 3, lane 2). In another set of experiments where we adjusted the amount of S1 nuclease, there appear to be three clusters of S1-sensitive segments: S-S1 (strong S1-sensitive segment; ϳ171 to ϳ197), which is most sensitive to S1 nuclease; W-S1 (weak S1-sensitive segment; ϳ212 to ϳ264), which is less sensitive; and E-S1 (extended S1-sensitive segment; from ϳ170 toward 5Ј-proximal), which is only weakly sensitive (Fig. 4, lane 2). They were detected only on the heattreated and not on untreated duplex DNA (Fig. 4, lanes 1 and 2). The S1 nuclease cleavages were observed on the wt Rif1BS I:2663 DNA but not on the Rif1CS mutant sequence (mut6 containing GGG to AGA mutations at all of the relevant G-tracts; Fig. 3A, lanes 5 and 6). On the mutant DNA fragment, two strong cleavage sites were detected irrespective of heat treatment. This may be caused by some specific DNA sequence/structure intrinsic to this mutated DNA fragment. The introduced mutation in mut6 created multiple copies of AG that could generate triplex or hairpin structures in the template DNA (40). The appearance of E-S1 was more variable between different sets of experiments.
We also examined the sensitivity of the same fragment to DNase I. Rather uniform DNase I cleavages were observed on the nontreated duplex DNA, whereas DNase I-resistant segments were detected on the heat-treated DNA. There appear to be two DNase I-resistant segments: S-DN (strongly resistant to DNase I; from ϳ161 to ϳ215), which is strongly protected, and W-DN (weakly resistant to DNase I; from ϳ216 to ϳ300), which is less strongly protected from DNase I attack (Figs. 3 (lane 4) and 4A (lanes 14 and 18)). All of the DNase I-resistant segments were detected only on the heat-treated/reannealed DNA fragment and not on the untreated dsDNA (Fig. 4A, lanes 13, 14, 17, and 18). They were not clearly observed on Rif1BS I: 2663 mut6, although slight reduction of DNase I sensitivity was observed in heat-treated mut6 (Fig. 3, lanes 7 and 8). These results indicate that the generation of S1 nuclease-sensitive and DNase I-resistant higher-order structures on Rif1BS I:2663 depends on the presence of the multiple G-tracts and heat denaturation/reannealing of duplex DNA. These results provide strong evidence for the formation of G4-like structures on the duplex Rif1BS. S1 sensitivity mapping results indicate the presence of unusually long (23-32 nt in Rif1BS I:2663 and 51 nt in Rif1BS II:4255 ) loop DNAs that connect the G-tracts in the proposed G4 structure.
We previously showed that the 3G-tracts in addition to Rif1CS are also important for formation of G4 structures in the ssDNA and also for in vivo chromatin binding of Rif1 (25). Therefore, we examined the effect of mutations (mut4) of the three 3Gs alone near the first Rif1CS I-1 (3G I-1 , 3G I-2 , and 3G I-3 ) of Rif1BS I:2663 on formation of higher-order structure in duplex DNA. The S1 nuclease sensitivity on S-S1 and E-S1 was reduced or exhibited altered cleavage patterns in mut4, in which three 3Gs were replaced by AGA, whereas that on W-S1 was largely unaffected with slight changes of cleavage patterns (Fig. 4A, compare lanes 2 and 4 or lanes 6 and 8). It is of interest that the segment spanning 3G I-1 and 3G I-2 is completely refractory to S1 digestion in the wt but is extensively digested in mut4 at a lower S1 concentration (indicated by pink bars on lane 4 of Fig. 4A), suggesting that this segment serves as a loop in the alternative G4 in mut4. We propose that mutation in 3Gs induced alternative G4 structures utilizing the

G-quadruplex structures on a duplex DNA
2G sequence present nearby. The W-S1 segment extended up to Rif1CS I-1 in mut4. We speculate that the involvement of 6G at Rif1CS I-1 in G4 formation with 3G I-4 and Rif1CS I-2 interfered with its involvement with guanines near 3G I-1 and 3G I-2 (Fig. 4C).
The S-DN was less protected from DNase I attack in mut4 compared with the wt sequence. Interestingly, the adjacent weakly protected segment (W-DN) became slightly more strongly protected in mut4 (Fig. 4A, compare lanes 18 and 20). This indicates that these 3Gs are also important for generation of a higher-order structure on this duplex Rif1BS, especially in the segment spanning the S-S1/S-DN, and in mut4, alternative higher order structures may be generated.
In one of the S1 nuclease assays, we observed additional S1 nuclease cleavages in the segment 5Ј to E-S1. This sensitivity is observed only in the heat-treated DNA (Fig. 4A, lanes 1 and 2), suggesting the presence of some additional secondary structures in this segment. However, we did not study the structure of this segment any further.
We also used T7 endonuclease to probe the structures generated on double-stranded Rif1BS I:2663 . We detected additional T7 endonuclease-sensitive sites upon heat denaturation of the . S1 nuclease sensitivity and DNase I resistance of duplex Rif1BS DNA depend on heat denaturation of the template DNA and the presence of G-tracts. The dsDNA from Rif1BS I:2663 (5Ј-end-labeled on the G-strand; wt or mut6 (mutations in all of the relevant G-tracts); 10 fmol) was heat-denatured and reannealed in the presence of 50 mM KCl and 40% PEG 200 (ϩ) or not treated (Ϫ), followed by S1 nuclease (1 unit) or DNase I (diluted 1:3000) treatment. The samples were analyzed on 6% PAGE containing 8 M urea. Marker lanes are DMS/piperidine products of the fragments shown. Red and blue lines show, respectively, the S1 nuclease-sensitive and DNase I-resistant segments that appear only after heat treatment and in the wt Rif1BS but not in the mutant that carries mutations in the G-tracts. The two strong S1-sensitive sites in the mutant template appear without heat treatment and are caused probably by a secondary structure on this DNA.

G-quadruplex structures on a duplex DNA
template DNA (Fig. S4, compare lanes 1 and 2 and lanes 3 and  4). These heat-induced sensitive sites are clustered in the vicinity of the Rif1CS I-1 , 3G I-1 , and 3G I-2 and appear to be located between the G-tracts or at the edge of the G-tracts. In contrast, the pattern of T7 endonuclease sensitivity did not change with the mutant (mut6; containing GGG to AGA mutations at all of G-quadruplex structures on a duplex DNA the relevant G-tracts) Rif1BS I:2663 even after heat treatment (Fig. S4, compare lanes 7 and 8 and lanes 9 and 10). Two strong sensitive sites were detected on the mutant DNA at locations very similar to those mediated by S1 nuclease.
These results are in support of the formation of G4 structures on Rif1BS duplex DNA despite the unusual arrangement of G-tracts. They indicate the presence of long loop segments connecting G-tracts. They also suggest that alternative structures are formed upon mutation of a part of the G-tracts, as was observed with single-stranded G-strand DNA sequences derived from Rif1BS (Fig. S2).

Effect of monovalent cations on the formation of higher-order DNA structures
We examined the effect of monovalent cations on the formation of these higher-order structures. It has been well established that formation of G4 structures requires the presence of a monovalent cation, such as K ϩ or Na ϩ , and that Li ϩ generally cannot support G4. Although Rif1BS I:2663 DNA generated in the presence of KCl was sensitive to S1 nuclease, that generated in the presence of LiCl exhibited much reduced sensitivity ( Fig.  S5A, lanes 1 and 2). Similarly, Rif1BS II:4255 heat-denatured in the presence of KCl exhibited one strong cleavage site between the 3G II-1 and Rif1CS II-1 and the clusters of cleavages from ϳ210 to Rif1CS II-2 , but the same DNA heat-denatured in the presence of LiCl was not digested by S1 (Fig. S5A, lanes 5 and 6). The sequences from Rif1CS II-1 to ϳ210 are a part of a stable G4 structure (see Fig. S5B) and thus are likely to be resistant to S1 nuclease. The strong S1-sensitive site upstream of Rif1CS II-1 may reflect the presence of a short loop between Rif1CS II-1 and 3G II-1 in the generated G4 structure. In contrast, mutant (mut3; both Rif1CS mutated) Rif1BS I:2663 or Rif1BS II:4255 did not show any heat-inducible S1 nuclease sensitivity in the presence of either KCl or LiCl (Fig. S5, A and B, lanes 3, 4, 7, and 8). These results gave further support to the idea that the structures generated on the G-rich strands from double-stranded Rif1BS DNAs adopt G4-like structures (Fig. S5B).

Higher-order structures are generated on the other strand of Rif1BS duplex DNA as well
The above results indicate that the G-rich strand (referred to as "G-strand") adopts a higher-order structure reminiscent of G4 upon heat denaturation and reannealing of the duplex DNA containing Rif1BS. We next examined by similar nuclease mapping whether the other strand of Rif1BS I:2663 adopts a higherorder structure. We end-labeled the other strand ("C-strand") of the duplex Rif1BS I:2663 and treated it with DNase I or S1 nuclease after heat treatment. In S1 nuclease digestion, clusters of S1-mediated cleavages were observed only on the heat-denatured wt DNA and not on the similarly treated mutant DNA in which two Rif1CS were mutated (Fig. 5, A and B, lanes 1-4).
There are two clusters of S1 nuclease-sensitive segments, one strong (ϳ238 to ϳ280) and the other weak (ϳ175 to ϳ225) (Fig.  5, A and B (lanes 1 and 2) and C). Thus, S1 nuclease sensitivity on the C-strand depends on the presence of the higher-order structures on the G-strand or on the presence of the C-tracts on the C-strand.
The C-strand of Rif1BS II:4255 also exhibited nuclease sensitivity upon heat denaturation (Fig. S6, lane 2). The sensitive sites were located over the segments between the clusters of the G-rich sequences and Rif1CS II-2 . T7 endonuclease-sensitive sites were also detected on the C-strand of Rif1BS I:2663 (Fig. S7,  lanes 3, 4, 7, and 8). Strongly sensitive sites were detected over the segments from Rif1CS I-1 to Rif1CS I-2 , and weakly sensitive sites were detected in the segments on both its upstream and downstream. In contrast to the S1 nuclease digestion, the segment between Rif1CS I-1 and 3G I-1 /3G I-2 was relatively insensitive to T7 endonuclease.
Because there are G-tracts on the C-strand of Rif1BS I:2663 , we next explored the possibility that this strand also adopts a G4-like structure. We took the 95-nt sequence of this strand (in Rif1BS I:2663 ) encompassing the S1 nuclease cleavage cluster (Fig. S7B). We also prepared a mutant version of this sequence in which GG, GGG, and GGGG were replaced with AG, AGA, and AGAG, respectively. The wt sequence generated slowly migrating forms on PAGE, whereas the mutant sequence did not under the same conditions (Fig. S8, lanes 1, 5, and 9). The BG4 antibody bound to the wt sequence but not to the mutant (Fig. S8, lanes 13-18 and 19 -21). These results indicate that the C-strand is also capable of forming G4.
Furthermore, Rif1 protein bound to the wt but not to the mutant sequence in gel shift assays (Fig. S8, lanes 2-4, 6 -8, and 10 -12). These results suggest that Rif1 could interact not only with the G4 structures generated on the G-strand but also with that generated on the C-strand. Thus, the duplex Rif1BS sequence may adopt complex higher-order structures that may contain G4-like structures on both strands.
Next, we generated the same mutations in Rif1BS I:2663 sequence (mut5; mutations in the G-tracts on the C-strand) and conducted S1 nuclease and DNase I sensitivity assays on both G-and C-strands. On the C-strand, a segment between Rif1CS I-1 and G3 (one of the G-tracts on the C-strand) was strongly protected from DNase I digestion (Fig. S9A, lane 14, blue solid line). This segment overlaps with the S1 nucleasesensitive segment (Fig. S9A, lanes 2 and 6, red line). We also detected additional protection from DNase I digestion over G1-G3 and further upstream (Fig. S9A, lane 14, blue dotted line). The former protection is not affected by the mut5 mutation, but the latter protection in the upstream segment appears to be slightly reduced in mut5 (Fig. S9A, compare lanes 14 and  16). We also noted localized changes of S1 nuclease and DNase (5Ј-end-labeled on the G-strand; wt or mut4 (mutations of the three GGG sequences to AGA); 10 fmol) was nontreated (Ϫ) or heat-denatured/reannealed (ϩ), followed by S1 nuclease (1 or 3 units) or DNase I (diluted 1:5000 or 1:3000) treatment. The samples were analyzed on 6% PAGE containing 8 M urea. Lanes 9 -12, DMS/piperidine products of the fragments shown. B, summary of the results showing the S1-sensitive and DNase I-resistant segments on Rif1BS I:2663 . C, induction of alternative G4 structures in mut4. Red brackets indicate the locations of S1 cleavage (i.e. loops in G4). Upon loss of 3G-tracts (light blue boxes; mut4), different loop segments are observed. 2Gs near 3G I-1 and 3G I-2 , together with Gs in the upstream segment, may be utilized for alternative G4 in the mut4 mutant.

G-quadruplex structures on a duplex DNA G-quadruplex structures on a duplex DNA
I sensitivities on the G-strand. A weak S1 nuclease cleavage near 3G I-1 was reduced (Fig. S9B, lanes 6 and 8), and DNase I sensitivity increased near C1-C3 (G-tracts on the C-strand) in mut5 on the G-strand (Fig. S9B, lanes 14 and 16), suggesting that the C-strand sequence could also affect the higher-order structure of the G-strand. However, the overall higher-order structure of Rif1BS does not appear to be affected by the G-tract mutations on the C-strand, although they might affect the local and fine structure of Rif1BS.

The effect of 7-deazaguanine on the generation of higherorder DNA structures on Rif1BS duplex DNA
The formation of G4 depends on Hoogsteen base pair, which utilizes the N7 position of the purine base. The replacement of guanines in the G-quartet with 7-deazaguanines results in disruption of G-quartet structure, thus loss of G4. We examined the effect of 7-deazaguanines on the formation of higher-order DNA structure on the duplex Rif1BS. For that, we generated a 180-bp DNA segment derived from Rif1BS I:2663 containing both Rif1CS I-1 and Rif1CS I-2 (see "Experimental procedures" and Fig. S10). S1 digestion after heat denaturation/reannealing of this fragment resulted in digestions at sites similar to those observed on the longer fragment, mainly in the loop segment between 3G I-2 and Rif1CS I-1 on both strands and weakly in the loop between 3G I-3 and 3G I-4 on the G-rich strand (Fig. 6). The appearance of the S1-sensitive sites depended on prior heat denaturation/reannealing. S1 cleavages were observed in the presence of KCl but not LiCl (Fig. 6, A and B, compare lanes 5 and 6 and lanes 7  and 8). The mutant DNA fragment in which all of the 3G-tracts and both Rif1CS were mutated did not show any S1 sensitivity even after heat denaturation/reannealing (Fig. 6, A and B, compare lanes 1 and 2 and lanes 5 and 6; see also Fig. S11, A and B). All of these data are consistent with those obtained with longer Rif1BS I:2663 duplex DNA. We noted that the G-tracts on the G-strand at Rif1CS I-1 and at 3G I-1 and 3G I-2 are almost completely protected from S1 digestion. This is consistent with the prediction that these G-tracts are involved in the four-stranded structure of G4 and are refractory to S1 attack. On the other hand, digestions extend more or less uniformly over the complementary C-tracts on the C-strand. This indicates that the higher-order structure, if any, adopted by this strand may not involve the C-tracts complementary to the G-tracts.
We next tried to see the effect of 7-deazaguanines on the higher-order structures. To prepare the templates containing 7-deaza-dGTP, the single-stranded regions were filled in on the partially heteroduplex generated with the two 100-mer ssDNAs, replacing guanines with 7-deazaguanines on the 3Ј 80 nt of both strands (see "Experimental procedures" and Fig.  S10A). The S1 nuclease-sensitive sites located between 3G I-2 and Rif1CS I-1 were greatly reduced on this 7-deazaguaninereplaced DNA template after heat treatment. The weaker S1-sensitive sites between 3G I-3 and 3G I-4 were also reduced on the deazaguanine template (Fig. 6, C and D, compare lanes 11  and 12 and lanes 15 and 16). On the 180-bp DNA fragment, we observed additional weak S1 cleavages 5Ј to 3G I-1 /3G I-2 on the G-strand, and these cleavages were not significantly affected by the presence of 7-deazaguanine. To specifically determine the effect of the G-strand deazaguanines, we generated the 180-bp DNA carrying 7-deazaguanines only on the G-strand 3Ј 80 nt (see "Experimental procedures" and Fig. S10A). The S1-sensitive sites disappeared in a very similar manner on this DNA, suggesting that the effect is mainly caused by the 7-deazaguanines on the G-strand (Fig. 6C, compare lanes 1 and 2 and lanes  5 and 6).
The results suggest that the guanines on the C-strand are not essential for the formation of stable higher-order structures on the Rif1BS duplex DNA. Indeed, S1 sensitivity on the C-strand was not affected by 7-deazaguanine substitutions of the 3Ј 80 nt of the C-strand (Fig. 6D, compare lanes 1 and 2 and lanes 5 and  6). On the other hand, the deazaguanine replacement of the G-strand greatly reduced the S1 sensitivity on the C-strand between 3G I-2 and 3G I-4 (Fig. 6D, compare lanes 11 and 12 and  lanes 15 and 16).
These results indicate that 7-deazaguanine replacement of guanines in the 3Ј-portion of the G-strand (3G I-3 , 3G I-4 , and Rif1CS I-2 ) results in loss of the S1 sensitivity in almost the entire segment or significant reduction/alteration of cleavage patterns, suggesting that these G-tracts contribute to the formation of the stable G4 structure on the duplex Rif1BS sequence. They also suggest the formation of alternative structures upon deazaguanine replacement of portions of G-tracts, consistent with the earlier results with the G-tract mutations on singleand double-stranded Rif1BS sequences. The results also support the idea that the higher-order structure on the C-strand is formed in a manner dependent on the formation of G4 structure on the G-strand.

The structure of Rif1BS and interaction with Rif1 revealed by polymerase stop assays
We have prepared Ͼ400-nucleotide-long ssDNA fragments containing either strand of Rif1BS I:2663 (see "Experimental procedures") and conducted polymerase stop assays after heat denaturation/annealing (Fig. S12A). The DNA chain elongation is expected to be stalled upon encounter with G4 on the template DNA. After DNA chain elongation by the Klenow fragment, a strong stop signal was detected on the G-strand of the wt Rif1BS 3Ј to the G-tract in Rif1CS I-2 (at positions ϳ283 to ϳ285) (Fig. S12B). Additional stops were identified at positions ϳ212 to ϳ214, immediately 3Ј to the 3G I-3 , which are located at the 3Ј-boundary of the S1 nuclease-sensitive segment. In contrast, on the mutant G-strand, stop signals appeared at many different locations, probably reflecting local secondary structures or alternative higher-order structures Figure 5. S1 nuclease mapping on the C-strand of duplex Rif1BS I:2663 . The duplex Rif1BS I:2663 DNA (5Ј-end-labeled on the C-strand; wt or mut3 (mutations in both Rif1CS); 10 fmol), heat-denatured and reannealed in the presence of 50 mM KCl and 40% PEG 200, was digested with S1 nuclease (lanes 1 and 3, 1 unit;  lanes 2 and 4, 3 units). The samples were analyzed on 8% (A) or 6% (B) PAGE containing 8 M urea. C, sequences of the C-strand of Rif1BS I:2663 DNA. S1 nuclease-sensitive segments are underlined. Sequences highlighted by a gray background were used for analyses in Fig. S8. Relevant G-tracts on the G-strand (indicated by C) and on the C-strand (indicated by G) are shown in blue and orange, respectively.

G-quadruplex structures on a duplex DNA
formed by the mutant ssDNA, and longer DNAs including the fully elongated DNA were also detected (Fig. S12, lanes 3 and 4). The introduced mutations generated highly AT-rich sequences which could generate a secondary structure, stalling the movement of DNA polymerase.
These results are consistent with the formation of a G4-like DNA structure on the Rif1BS sequence that impedes the movement of DNA polymerase. The two eminent stop sites coincide with the edge of S1 nuclease-sensitive/DNase I-resistant segments (see Fig. 8), consistent with the idea that two G4-like structures are formed on the DNA. On the C-strand, clear stop signals were not detected on either wt or mutant sequences (Fig. S12B, lanes 5 and 6), suggesting the absence of a stable secondary structure that would inhibit the movement of DNA polymerase on this strand under the conditions employed.  3, 4, 7, and 8) and was digested with S1 nuclease (lanes 1, 3, 5, and 7, 3 units; lanes 2, 4, 6, and 8, 9 units). Lanes 1-4, mut3 (mutations in all of the G-tracts); lanes 5-8, wt. The locations of two G-tracts from Rif1CS and four G3-tracts are shown, respectively, with green and blue lines along the gel. C and D, regular and 7-deazaguanine-replaced 180-bp duplex Rif1BS I:2663 DNAs (5Ј-end-labeled on the G-strand (C) or the C-strand (D), heat-denatured, and reannealed (lanes 1, 2, 5, 6, 11, 12, 15, and 16) or nontreated (lanes 3, 4, 7, 8, 13, 14, 17, and 18) in the presence of 50 mM KCl and 40% PEG 200) were digested with S1 nuclease (lanes 1, 3, 5, 7, 11, 13, 15, and 17, 3 units; lanes 2, 4, 6, 8, 12, 14, 16, and 18, 9 units). Lanes 5-8, only the 3Ј 80 nt on the labeled strand were replaced with 7-deazaguanines. Lanes 15-18, the 3Ј 80 nt on both strands were replaced with 7-deazaguanines. The 7-deaza-replaced segments are shown in red. The end-labeled DNA is blue. The samples were analyzed on 8% PAGE containing 8 M urea in A-D. E, summary of S1-sensitive sites on the 180-bp duplex DNA derived from RIf1BS I:2663 . S1-sensitive segments on the G-strand (top strand) or the C-strand (bottom strand) are shown. Blue and black lines below each strand represent S1-sensitive segments on the normal and deazaguanine-replaced (on the 3Ј-half of the G-strand) templates, respectively. Relevant G-tracts on both strands are shown in large capital letters. Residues in boldface type (and the underlined segment in the sequences of C and D) are the annealing segment of two oligonucleotides, and therefore, guanine in this segment cannot be replaced with deazaguanines. Two G-tracts from Rif1CS and four G3-tracts are boxed in green and blue, respectively. See Fig. S11 for scanning of the bands in the gels.

G-quadruplex structures on a duplex DNA i-motif formation is not likely to be involved in higher-order DNA structure on the C-strand
It has been argued that G4 formation on duplex DNA may induce the generation of the i-motif structure (34,(41)(42)(43), composed of two parallel duplexes with intercalated hemiprotonated cytosine ϩ -cytosine (C ϩ -C) bp. We examined the potential contribution of the i-motif structure to the formation of a higher-order DNA structure on the duplex Rif1BS. The pH of the buffer for heat denaturation was lowered to pH 5.8, optimal for i-motif formation. In this experiment, KCl was not included, and thus G4 formation would not occur. Under these conditions, S1 nuclease sensitivity was completely gone on both strands even after heat treatment (Fig. 7A, lanes 3, 6, 9 , 12, 15, 18, 21, and 24). We then heat-treated DNA in the pH 5.8 buffer containing only 50 mM KCl (without PEG 200). Unexpectedly, the S1 nuclease sensitivity and DNase I protection were completely lost on both strands under these conditions as well (Fig.  7B, compare lanes 1 and 3, lanes 5 and 7, lanes 9 and 11, and  lanes 13 and 15). These results are in support of the idea that G4 formation facilitated by KCl and the molecular crowding effect is the major driving force for generation of higher-order DNA structures at Rif1BS on both DNA strands and that contribution of the i-motif structure is very limited under the experimental conditions employed. Alternatively, i-motif formation on the C-strand, if any, completely depends on the G4 formation on the G-rich strand.

Discussion
Sequences capable of forming G4 are ubiquitous on genomes (2,6,7,9,44,45). These sequences can be formed in vitro on ssDNA sequences, commonly described as ϭ 1-7). Although the structures of G4 generated from ssDNA have been studied in detail, those of G4 generated on duplex DNA have been elusive. To elucidate the roles of G4 in vivo, it would be important to clarify the structures of G4 generated on duplex DNA. We reported genetic evidence that strongly indicates that a conserved nuclear protein, Rif1, binds to chromatin in nuclei by recognizing G4 structures and regulates temporal and spatial program of DNA replication (25,30). Therefore, in this report, we have used the binding target sequence of Rif1, Rif1BS, as a model to analyze their structures

G-quadruplex structures on a duplex DNA
in a duplex form, when forced to form higher-order structures by heat denaturation/reannealing. The results revealed the following new information on the structures of the G4-forming Rif1BS DNA. 1) G4 is formed on the G-strand on duplex Rif1BS DNA by heat denaturation/reannealing. 2) The G4 generated on the duplex Rif1BS contains unusually long loops that are sensitive to S1. 3) The formation of a G-quadruplex on the G-strand is associated with that of a higher-order DNA structure on the other strand (C-strand). 4) This structure is formed in a manner dependent on the G-tracts on the G-strand and is not an i-motif, as was previously anticipated. 5) G-quadruplex structures formed on Rif1BS are robust in that they can form alternative structures upon loss or inactivation of some G-tracts involved in the G4 formation.

ssDNAs derived from Rif1BS could form oligomeric G4 structures, which are the preferred structures for Rif1 binding
Rif1BS often carries two Rif1CS harboring 5G-or 6G-tracts, which are separated by ϳ100 bp. Additionally, 3G-or 4G-tracts are present in the vicinity of the 5/6Gs. The distance between the 5/6G and the nearest 3G (shown to contribute to the G4 formation) can be as far as 49 bp. This is quite unusual, consid-ering the previous view of the canonical G4 structures. Various ssDNA sequences, derived from Rif1BS I:2663 and Rif1BS II:4255 , upon heat denaturation, form a variety of structures that migrate slowly on PAGE (Fig. 1). Two conclusions can be drawn from this set of experiments. 1) ssDNA sequences containing either or both of the two Rif1CS can form slowly migrating forms probably composed of multiple DNA molecules on a polyacrylamide gel. 2) Rif1 can bind to these slowly migrating forms but not to the fast migrating forms, probably monomer G4.
Rif1-11, an 81-nucleotide sequence derived from Rif1BS II:4255 , could form three major and other minor slowly migrating forms. The major forms are monomer, dimer, and probably tetramer of a G4 moiety. Generation of these forms depends on the 5Ј-proximal G-tract. Loss of the 5Ј-proximal G-tract eliminated both of the major forms, but large aggregates at the well remained. On the other hand, the loss of 3Ј-proximal G-tract eliminated only one of the major slowly migrating forms, presumably a tetramer, that is bound with Rif1 most efficiently (Fig. 2). The DMS footprint indicated that slightly different G4 structures are formed in the latter mutant (mut1). Thus, it  1-12) or C-strand (lanes 13-24); either wt or mut5; 10 fmol) were heat-denatured/reannealed under normal conditions (with 50 mM KCl and 40% PEG 200; ϩ) or in a 100 mM cacodylate buffer (pH 5.8) without KCl (red; ϩ). In 50 mM KCl and 40% PEG 200, one sample was not heat-treated as a control (Ϫ). DNA was treated with S1 nuclease (1 or 3 units) and analyzed on a denaturing gel. B, the same DNA fragments (5Ј-end-labeled on the G-strand (lanes 1-8) or C-strand (lanes 9 -16); either wt (w) or mut5 (m); 10 fmol) were heat-denatured/reannealed and treated with S1 nuclease (3 units) or with DNase I (1:3000 dilution). DNA was heat-denatured at a neutral pH with 40% PEG 200 and 50 mM KCl (under which G4 can be formed; lanes 3,4,7,8,11,12,15,and 16;Normal) or in 100 mM cacodylate buffer (pH 5.8) and 50 mM KCl (under which i-motif can be formed; lanes 1, 2, 5, 6, 9, 10, 13, and 14; pH 5.8). The samples were analyzed on 6% PAGE containing 8 M urea. The end-labeled DNA was restriction-digested (PvuI for G-strand and MluI for C-strand), and the longer fragments were reisolated from PAGE, thus resulting in different lengths for the two fragments.

G-quadruplex structures on a duplex DNA
appears that the Rif1-11 sequence has a potential of forming multiple higher-order structures, some stable and others unstable, depending on which G-tracts (of Rif1CS) are available (Fig.  S2). It is not clear why Rif1 binds preferentially to high-molecular weight forms (oligomers) of G4. We have observed that G4 can self-associate in vitro and Rif1 can promote the self-association. 4 We speculate that multiple G4s present over a stretch of the chromatin segment may self-associate to generate "oligomeric states," which Rif1 may preferentially bind to and stabilize. This may contribute to generation of a replication-suppressive chromatin domain through chromatin loop formation.

Higher-order DNA structures are generated on both strands in duplex Rif1BS
In vitro, binding of Rif1 to duplex DNA is nonspecific, and specificity to Rif1BS is apparent only after duplex DNA is denatured/reannelaed by heating (25). DNA species, which migrate slowly on PAGE, are generated upon heat treatment of Rif1BS duplex DNA. The formation of this DNA structure requires N7 of guanine and is stabilized by a G4 ligand, suggesting generation of a G4-related structure (25). To prove this prediction, we used nucleases to probe the structures generated on duplex Rif1BS after heat denaturation. DNase I and S1 nuclease were used previously to investigate the structure generated at a G-rich sequence in the VEGF promoter on a supercoiled plasmid DNA (46). It was reported that the G-rich segments were protected from digestion by both nucleases in the presence of KCl. Our nuclease mapping data show that a ϳ120-bp segment on the G-strand is protected from DNase I digestion after heat treatment and that this protection is not observed in a mutant that has lost G-tracts in Rif1CS (Fig. 8), suggesting the presence of a higher-order structure, most likely G4-like structures. Con-sistent with this prediction, extensive S1 nuclease sensitivity was observed on the G-strand. S1 nuclease cleavages and protection from DNase I attacks were detected mainly on the segments between the G-tracts, consistent with the presence of loop structures predicted for the G4 structure. Interestingly, one strong and one weak S1 nuclease-sensitive site (S-S1 and W-S1) were detected on Rif1BS I:2663 in a manner dependent on the G-tracts. A similar long stretch of S1-sensitive segment was detected on Rif1BS II:4255 between 3G II-2 and Rif1CS II-2 (Fig.  S5). S-S1 and W-S1 correspond to the segments strongly and weakly protected from DNase I attack, respectively (S-DN and W-DN). The S-S1 is flanked by Rif1CS I-1 and 3G I-1 /3G I-2 . We showed previously by DMS footprint and a nuclease sensitivity assay that the ssDNA covering this segment adopts G4 structure in a manner dependent on the 3Gs and the long G-tract in Rif1CS I-1 (25). The W-S1 is present between the 3G I-3 and another long G-tract within the Rif1CS I-2 . Polymerase stop assays show two major stop sites: one at Rif1CS I-2 (near the boundary of W-S1/W-DN) and the other at the 3G I-3 (near the boundary of S-S1/S-DN; see Fig. 8). Thus, at least two G4like structures may be formed on Rif1BS I:2663 by involving the 120-bp segment containing two Rif1CS and other GGG sequences. This is supported by the fact that the mutations of the three GGG sequences (3G I-1 , 3G I-2 , and 3G I-3 ; mut4) result in the loss of S-S1 (as well as of protection from DNase I attack at S-DN), but not in loss of W-S1 (or of protection at W-DN), whereas the mutations of both Rif1CS result in the loss of both S-S1 and W-S1 and of protection from DNase I attack at both S-DN and W-DN. Loss of three GGG sequences appears to induce different higher-order structure at the distal W-DN/ W-S1 segment, as judged by altered nuclease sensitivity (Fig. 4). Interestingly, the W-DN segment was slightly more protected from DNase I digestion, and stronger S1 cleavages were detected at localized positions on W-S1 in mut4 than in the wt   (Fig. 4A). These could suggest that the G4 structures generated on the W-DN/W-S1 segment in mut4 of Rif1BS I:2663 could be slightly more stable than that in the wt. Alternative G4 structures induced by G-tract mutations are consistent with the effect of G-tract mutations on the structure of Rif1-11 ssDNA derived from Rif1BS II:4255 ( Fig. 2 and Fig. S2). S1 nuclease digestion of the other strand (C-strand) of Rif1BS I:2663 showed the presence of sensitive sites on the sequences complementary to the S1 nuclease-sensitive segments on the G-strand, most notably those complementary to the S-S1. We noted the presence of G-tracts in this segment, and indeed this sequence appears to be capable of forming a G4 structure (Fig. S8). Therefore, G4 structures may be generated on both strands on a heat-treated Rif1BS duplex DNA. However, the mutations of the G-tracts on the C-strand did not significantly affect the overall higher-order DNA structures of the heat-treated duplex DNA or compromise the Rif1 binding to chromatin in vivo (25) (data not shown), although we detected some local changes of nuclease sensitivities (Fig. S9). Replacement of the guanines on the C-strand with 7-deazaguanine did not affect the S1 sensitivity on this strand (Fig. 6D). Thus, the secondary structures on the C-strand does not appear to depend on its potential ability to form G4 at least on Rif1BS I:2663 . However, 25% of the 35 strong Rif1BS contain Rif1CS (carrying 5/6G-tracts) on both strands (25), and thus, it is still possible that G4 structures are formed on both strands at these Rif1BS.

Higher-order DNA structures at Rif1BS are mainly driven by G4, not i-motif
Although the presence of C-rich sequence predicts the formation of i-motif on the C-strand, our experimental conditions (pH 7.5) do not permit it. Lowering the pH did not facilitate the formation of the higher-order structure but instead destroyed it (Fig. 7). It was previously reported that on a duplex DNA containing G4-forming sequence, only one of the structures, G4 or i-motif (but not both), is formed (47,48). Generally, G4, which is thermodynamically more stable than i-motif, is preferentially formed. Heat denaturation of duplex Rif1BS under conditions where i-motif could be formed but G4 may not be stably formed (low pH (pH 5.8) in the absence of PEG 200) did not generate any S1 nuclease-sensitive sites or protection from DNase I attack on the C-strand or on the G-strand (Fig. 7B). The polymerase stop assay on the C-strand did not show significant stop signals (Fig. S12), suggesting that the structure generated on the single-stranded form of the C-strand, if any, is not as stable as the one synthesized on the G-strand. These results support the notion that the higher-order structure that may be present on the C-strand of Rif1BS I:2663 is generated passively as a result of G4 formation on the G-strand, although the G4-forming potential on the C-strand may facilitate this process in some cases. We noted that S1 cleavages were exclusively on the sequences between the G-tracts and absent on the G-tracts on the G-strand, consistent with the prediction that guanines are involved in four-stranded structure. On the other hand, on the C-strand, cleavages occur also on the C-tracts, complementary to the G-tracts. This is consistent with the absence of i-motif and suggests the presence of an unknown higher structure on the C-strand.

Dynamic structural transition involving alternative G4 structures at Rif1BS and its biological implications
S1 nuclease sensitivity was detected also on the other strand of Rif1BS II:4255 , although in this case, we were not able to identify potential G4-forming sequences on this strand. Close examination of the 35 Rif1BS revealed that most of them carry many G-tracts on both strands, in addition to the 5G or 6G present in the Rif1CS. About 25% of the Rif1BS carry Rif1CS on opposite strands, suggesting the possibility that G4 could be formed on both strands.
We have arranged the 35 Rif1BS in order of the enrichment of ChIPed DNA (representing Rif1 binding efficiency; see supporting information for sequences of all of the Rif1BS). We have calculated the "G-tract scores" for both G-and C-strands of each Rif1BS and plotted them against the ChIP efficiency ranking. G-tract score is generally higher for the better Rif1BS (Fig.  S13A). The tendency is clearer with the C-strand. We selected the top six and bottom six Rif1BS and plotted the G-tract scores of both strands. Again, clustering of the top six or bottom six Rif1BS into the population with higher or lower G-tract score, respectively, is noted (Fig. S13B). Thus, the presence of long G-tracts on both strands may be related to better Rif1 binding, supporting the speculation that Rif1BS may form G4 on both strands. In a simplistic model, one can assume that more runs of G make more stable G4 (G4 made of GGG (three G-quartets) Ͻ G4 made of GGGG (four G-quartets) Ͻ G4 made of GGGGG (five G-quartets)). Thus, the numbers of G in G-runs could determine the stability of the G4 and affect the Rif1 binding. These results indicate the presence of multiple G-tracts on both strands as a general feature of Rif1BS and suggest the significance of the long tracts in generation of a higher-order structure that is favored by Rif1.
These general features of the Rif1BS sequences suggest that formation of multiple G4 structures on Rif1BS on one or both strands may be their common characteristics for efficient recognition by Rif1. We also propose, in conjunction with the robustness of the G4 formation on Rif1BS (see Figs. 1 and 4), that Rif1BS DNA may undergo conversion from one form to another, which may contribute to the dynamic nature of chromatin higher-order structures in cells. The dynamic structure transition of DNA could be a basis for the stochastic and plastic nature of cellular biological events (e.g. transcription initiation, replication initiation, and chromatin regulation in general).
Another revelation from this study is that G4 can be formed on the G-tracts with spacing quite diverged from the canonical G4-forming sequences ( . The spacing of G-tracts in Rif1BS II:4255 is as large as 49 nt, which becomes sensitive to S1 upon heat denaturation/reannealing. We probably need to consider using the algorithm that would more accurately reflect the G4 structure generated in vivo or even in vitro. In fact, only four Rif1BS of 35 overlapped with putative G4 motifs predicted to be present in the fission yeast genome (total of 446) (49). Recent DNA polymerase stoptype screening of genomic G4 sequences (G4-seq) revealed the G-quadruplex structures on a duplex DNA presence of more than 700,000 potential G4 on the human genome (50), where 375,000 G4 were previously predicted on the basis of the above canonical algorithm (6,7). These suggest that nucleic acids have the potential to form G4 more widely on genomes than previously anticipated.
Finally, dynamic transition between alternative higher-order DNA structures could be more common in chromosomal G4 structures. The genome-wide distribution of G4-forming sequences indicates that the G4 sites are not uniformly present on the genome (4) and are generally enriched in the transcription promoter segments, which are often GC-rich or a part of a GC-island and may be capable of forming multiple G4 structures. Thus, the findings in this report could provide novel insight into the structures of G4 in general.

Purification of fission yeast Rif1 protein
His-and FLAG-tagged Rif1 protein or its derivatives were expressed in 293T cells and purified as described previously (25,36).

DNA templates: Rif1BS-derived single-stranded and dsDNA
The sequences of the oligonucleotides used in the assays are described in each figure. All of the oligonucleotides, OPC column-purified, were further purified by denaturing polyacrylamide gel (containing 8 M urea) if their lengths exceeded 50 nt. Specifically, oligonucleotides, denatured by heating in 95% formamide and 5 mM EDTA, were run on 10% PAGE containing 8 M urea and eluted from the gel by crushing the gel pieces containing the DNA in elution buffer (0.5 M ammonium acetate, 10 mM magnesium acetate, 1 mM EDTA, and 0.1% SDS). After incubation at 37°C for 4 h, the oligonucleotide solution was passed through a Spin-X centrifuge tube filter (Corning), ethanol-precipitated, and resuspended in TE buffer.

Analyses of DNA-protein complexes on polyacrylamide gels
Labeled DNA fragments were mixed in reaction mixtures (10 l) containing 40 mM Hepes/KOH (pH 7.6), 50 mM KCl, 1 mM EDTA, 10 % glycerol, and 0.01 % Triton X-100, and purified proteins were added to initiate the binding reaction with purified proteins. In competition assays, labeled DNA fragment and cold competitor DNA were premixed before addition of the protein. After incubation at room temperature for 30 min, the reaction mixtures were directly applied onto a polyacrylamide gel prepared in 1ϫ TBE, 50 mM KCl, and 40% PEG 200.

DMS footprint mapping
An end-labeled DNA in 20 l of 60 mM cacodylate buffer containing 60 mM KCl was incubated with DMS (0.1-1 l) at room temperature for 5 min, followed by the addition of 1 l of 4 M ␤-mercaptoethanol and 80 l of 10% piperidine and incu-G-quadruplex structures on a duplex DNA bation at 96°C for 5 min. After digestion, 100 l of water containing 10 g of glycogen was added. DNA was ethanolprecipitated and washed with 70% ethanol. The pellet was resuspended in 6 l of deionized formamide containing 5 mM EDTA, heated at 96°C for 3 min, and applied onto polyacrylamide gel containing 8 M urea in 0.5ϫ TBE. The amount of DMS added was titrated in each experiment.

DNase I, S1 nuclease, and T7 endonuclease mapping
For DNase I digestion, 32 P-end-labeled DNA (heat-treated in 50 mM KCl and 40% PEG 200 or nontreated) was digested with DNase I (Takara; 5 units/l; 1 l of 1:1000 to 1:5000 dilution (0.005-0.001 units) was used per assay) in 10 l of 1ϫ DNA binding buffer containing 8 mM MgOAc and 0.5 mM CaCl 2 at room temperature for 2.5 min. For S1 nuclease mapping, DNA was digested with S1 nuclease (Thermo Fisher Scientific; 1-3 units) in 10 l of 10ϫ S1 buffer (supplied with the enzyme) at room temperature for 2.5-5 min. For T7 endonuclease mapping, DNA was digested in 50 mM NaCl, 10 mM Tris-HCl (pH 7.9), 10 mM MgCl 2 , and 1 mM DTT with T7 endonuclease I (New England Biolabs; 0.5-5 units) for 10 min at 37°C in 50 l. In all of the cases, the reaction was terminated by the addition of 200 l of 20 mM EDTA containing 10 g of glycogen, and DNA was ethanol-precipitated, washed, and resuspended in a small aliquot of formamide containing 5 mM EDTA. The amount of DNase I, S1 nuclease, and T7 endonuclease was titrated in each experiment.

DNA polymerase stop assay on ssDNA
Rif1BS I:2663 (wt and mut3) was amplified with T7-Rif1BS(ChrI) (CGGAATTCtaatacgactcactataggCGCATGTA-CGTATTTCTTTTTACA; T7 promoter sequence in lowercase letters) and Rif1bs-as-chr1(BssHII) (GTTGGCGCGCGGGT-AAGTCTATTCCGTCTCACTT) and cloned at the SmaI site of pUC18. The insert DNA was excised by EcoRI ϩ HindIII digestion and transferred to M13 mp18 and M13 mp19 vector using the same sites. Single-stranded phage DNAs were isolated from the recombinant phages produced in Escherichia coli JM103 cells. To isolate the insert ssDNA, the circular ssDNA from M13 mp18 was hybridized with the hybridizing primers mp18-EcoRI (GTACCGAGCTCGAATTCGTAATCATGGT-CAT; restriction site shown in italics) and mp18-HindIII (GCCAGTGCCAAGCTTGCATGCCTGCAG) to convert the restriction site segments to duplex and digested by EcoRI ϩ Hin-dIII. The released insert DNA (G-strand) was isolated from PAGE. The ssDNA from M13 mp19 was similarly hybridized with the hybridizing primers mp19-EcoRI (CGACGGCCAGT-GAATTCGAGCTCGGTACC) and mp19-HindIII (CTGCA-GGCATGCAAGCTTGGCGTAATCATG), and the insert ssDNA (C-strand) was isolated. The isolated ssDNA was calf intestinal alkaline phosphatase-treated and end-labeled with 32 P by T4 polynucleotide kinase. G-strand or C-strand ssDNAs were mixed with a 32 P-end-labeled chain-elongation primer, Rif1bs-s-chr1(MluI) (GCGACGCGTCGCATGTACGTATT-TCTTTTTACA) or Rif1bs-as-chr1(BssHII) (GTTGGCGCG-CGGGTAAGTCTATTCCGTCTCACTT), respectively; heated in the presence of 50 mM KCl and 40% PEG 200; and cooled down to room temperature. DNA chain was elongated from the 32 P-end-labeled chain-elongation primer by the Klenow fragment in 50 mM KOAc, 20 mM Tris acetate (pH 7.9), 10 mM Mg(OAc) 2 , 1 mM DTT, and 0.25 mM 4 dNTPs at 37°C for 60 min. The samples were heated in the presence of formamide and 5 mM EDTA and were applied on polyacrylamide gel containing 8 M urea in 0.5ϫ TBE.

CD spectrometry
The oligonucleotides were diluted to 2 M in the following buffers: 50 mM Tris-HCl without salt (pH 7.5), 50 mM Tris-HCl with 50 mM KCl (pH 7.5), and 50 mM Tris-HCl with 50 mM NaCl (pH 7.5). Subsequently, these solutions were denatured by heating at 99°C for 5 min and then slowly cooled to room temperature and incubated overnight. CD spectra were recorded on a J-720 spectropolarimeter (JACSO, Tokyo, Japan) using a quartz cell (Agilent; microcell 50 l, 10-mm optical path length) with an instrument scanning speed of 500 nm/min and a response time of 1 s over a wavelength range of 230 -320 nm. The CD spectra are representatives of five averaged scans taken at 25°C.