Cell Cycle Modulation of Gene Targeting by a Triple Helix-forming Oligonucleotide*

Successful gene-targeting reagents must be functional under physiological conditions and must bind chromosomal target sequences embedded in chromatin. Triple helix-forming oligonucleotides (TFOs) recognize and bind specific sequences via the major groove of duplex DNA and may have potential for gene targetingin vivo. We have constructed chemically modified, psoralen-linked TFOs that mediate site-specific mutagenesis of a chromosomal gene in living cells. Here we show that targeting efficiency is sensitive to the biology of the cell, specifically, cell cycle status. Targeted mutagenesis was variable across the cycle with the greatest activity in S phase. This was the result of differential TFO binding as measured by cross-link formation. Targeted cross-linking was low in quiescent cells but substantially enhanced in S phase cells with adducts in ∼20–30% of target sequences. 75–80% of adducts were repaired faithfully, whereas the remaining adducts were converted into mutations (>5% mutation frequency). Clones with mutations could be recovered by direct screening of colonies chosen at random. These results demonstrate high frequency target binding and target mutagenesis by TFOs in living cells. Successful protocols for TFO-mediated manipulation of chromosomal sequences are likely to reflect a combination of appropriate oligonucleotide chemistry and manipulation of the cell biology.

Effective gene-targeting reagents would have a broad application as probes of chromatin structure and in a variety of genomic manipulations, e.g. gene knock-out, targeted gene conversion and/or recombination, and gene therapy. Successful constructs must be functional in the nuclear environment of living cells. Furthermore, their intended targets must be accessible despite the packaging and condensation of nuclear DNA by chromosomal proteins. One targeting strategy that has been of interest for many years is based on triple-helix forming oligonucleotides (TFOs). 1 A DNA triple helix can form when a TFO lies in the major groove of intact duplex DNA (1)(2)(3)(4)(5). The most stable structures assemble on polypurine:polypyrimidine sequences with hydrogen bonds formed between the bases in the third strand and those in the purine strand of the duplex. The purine or pyrimidine motif third strands may be involved in triplex formation depending on the target sequence, and a binding code for the design of the third strands has been defined (6). Although conventional oligonucleotides have features that limit their activity under physiological conditions, there is a variety of chemical modifications that ameliorate these limitations (7). Thus, for pyrimidine TFOs, the replacement of cytosine by 5-methylcytosine and the use of 2Ј-O-methyl (2Ј-OMe)-modified sugars permit stable triplex formation at physiological pH in vitro (8 -13). Other derivatives contribute to the bioactivity of TFOs of both motifs as shown by us and others in recent publications (14 -18).
Although chemical modifications can enhance TFO affinity and triplex stability, they obviously cannot address the issue of access to chromosomal targets inside living cells. In biochemical experiments, it has been shown that sequences in nucleosomes are poor candidates for triplex formation (19 -22). Thus, in the cell, a TFO is likely to confront the same issues of target accessibility described for sequence specific regulatory proteins (23). Of course, reports of gene targeting in vivo by TFOs imply, of necessity, some degree of target access (14, 18, 24 -26). In addition, TFO binding to chromosomal targets in permeabilized cells has been shown (27,28).
We have employed an Hprt knock-out assay to measure the activity of TFOs linked to the DNA cross-linker psoralen (pso-TFOs). The assay reflects the binding of the target sequence by the TFO and, following photoactivation of the psoralen, the generation of an inactivating mutation as a result of cellular processing of the cross-link. We have used this approach to measure the activity of TFOs carrying sugar and base modifications and have shown that TFOs with 2Ј-O-aminoethyl (2Ј-AE) substitutions are bioactive (18). We have now asked whether the biological state of the cell could influence TFO activity. Here we show that the frequency of mutagenesis induced by a pso-TFO differs in quiescent and S phase cells. The difference reflects the levels of targeted cross-links in the two cell populations.

MATERIALS AND METHODS
Cells and Synchronization Protocol-Chinese hamster ovary (CHO) AA8 cells (ATCC) were grown in Dulbecco's modified Eagle medium (Invitrogen) supplemented with 10% fetal calf serum and penicillin and streptomycin. Prior to an experiment, the cells were grown in HAT medium (10 Ϫ4 M hypoxanthine, 5 ϫ 10 Ϫ5 M aminopterin, 10 Ϫ5 M thymi-* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. ¶ ¶ To whom correspondence should be addressed: Laboratory of Molecular Gerontology, NIA, National Institutes of Health, 5600 Nathan Shock Dr., Baltimore, MD 21224. Tel.: 410-558-8565; Fax: 410-558-8157; E-mail: seidmanm@grc.nia.nih.gov. 1 The abbreviations used are: TFO, triple-helix forming oligonucleotides; 2Ј-OME, 2Ј-O-methyl; pso-TFO, TFOs linked to the DNA crosslinker psoralen; 2Ј-AE, 2Ј-O-aminoethyl; CHO, Chinese hamster ovary; UVA, ultraviolet A; kb, kilobase. dine) for 1 week. The cells were synchronized in G 0 /G 1 by a variation of the method described by Sawai et al. (29). Cells were plated at subconfluent levels, and on the next day, the medium changed to Dulbecco's modified Eagle medium with 2% fetal bovine serum and 2% Me 2 SO. After 48 h, the cells were washed (ϳ85% G 0 /G 1 cells by fluorescenceactivated cell sorter analysis) and either electroporated or fed with complete medium (for G 1 phase experiments) or incubated with complete medium containing 100 M mimosine for 16 h to block them in early S phase (ϳ90% early S cells) (for complete details on this procedure, see Ref. 30). After 16 h, the cells were released from the mimosine block by feeding with Dulbecco's modified Eagle medium, 10% fetal bovine serum.
TFO Electroporation, Psoralen Treatment, and Hprt Mutation Assay-Cells were suspended at 10 7 /ml and mixed with TFO at 5 M. The cells were electroporated (Bio-Rad) followed by incubation at room temperature for 3 h and a 3-min exposure in a Rayonet chamber to UVA light at 1.8 J/cm 2 . The electroporation conditions were chosen to minimize cell toxicity (trypan blue staining showed ϳ95% viable cells after UVA treatment). Cells treated with free psoralen were incubated with 5 M psoralen for 30 min followed by UVA treatment. The cells were passaged and then exposed to thioguanine selection (17).
Restriction Resistance Analysis of Non-selected Clones-Following TFO electroporation, UVA treatment, and culture for 3-5 days to permit mutagenesis, 100 cells were plated in 60-mm dishes in standard growth medium. Individual colonies were expanded, and the DNA was extracted followed by amplification of the I4E5 target region and digestion of the PCR products with XbaI.
Cross-link Analysis-After UVA exposure of cells treated with the pso-TFO, the DNA was extracted and digested with EcoRI. 10 g in 10 l of solution were then mixed with 45 l of 98% formamide and heated at 77°C for 15 min. The samples were electrophoresed in an agarose gel in neutral buffer and transferred to a nylon membrane. The membrane was hybridized with a 2-kb EcoRI fragment containing the I4E5 target region and adjacent sequence. The probe was labeled by random priming and incubated overnight with a blank membrane in hybridization buffer to remove radioactive species that bound nonspecifically. This step greatly reduced nonspecific binding to the nylon. In other experiments, following TFO/UVA treatment, genomic DNA was exhaustively digested with EcoRI and XbaI, blotted, and hybridized. To control for triplex formation resulting from the interaction of unbound TFO and the target during DNA purification (31), we mixed AE-07 and cells and then isolated genomic DNA and digested it. There was no XbaI-resistant band.

RESULTS
The CHO I4E5 HPRT Target and pso-TFO-The triplex target sequence is in the fourth intron next to Exon 5 of the CHO Hprt gene (Fig. 1a). The sequence consists of a 17-base polypurine:polypyrimidine sequence ending in a 5Ј-TA step, which is appropriate for T-T interstrand cross-linking by psoralen (32). The A is the first base of the AG splice acceptor sequence, and point mutations at this site have been reported previously (33). The intron/exon junction also contains a recognition sequence for the restriction enzyme XbaI (TCTAGA). The TFO used in these studies, AE-07, was linked to psoralen and contained a patch of four 2Ј-AE substituted nucleotides, whereas the remainder of the molecule contained 2Ј-OMe sugars ( Fig. 1, a and  b). We have described the synthesis and characterization of TFOs containing 2Ј-AE substitutions in previous publications (17,18). The 2Ј-AE residues are protonated at physiological pH. They reduce the charge repulsion between the third strand and the duplex target, thus lowering the requirement for Mg 2ϩ (34,35). They also establish a stabilizing interaction with phosphates in the purine strand of the duplex target (36). Psoralenlinked TFOs containing the appropriate amount and distribution of this modification are active in the Hprt assay (18).
Pso-TFO Activity during the Cell Cycle-AE-07 was introduced by electroporation into quiescent CHO cells and into cells at different times after release from quiescence (see "Materials and Methods") (for fluorescence-activated cell sorter profiles of quiescent and mid S phase cells, see Fig. 2a). The psoralen was photoactivated, and the frequency of cells with inactivating mutations in the Hprt gene was determined via conventional thioguanine selection. The mutation frequency was quite low in the cells in G 0 /G 1 but increased in cells traversing G 1 and appeared maximal in S phase cells. To more precisely define the time of peak activity in S phase, the cells were released from G 0 /G 1 block and then blocked again in early S phase by treatment with complete medium containing mimosine (30). The cells were released from this block and, as before, were treated at various times with AE-07/UVA followed by determination of thioguanine-resistant colonies. The greatest activity was seen 4 -6 h after release during which time the cells were in mid-to-late S phase. Results from selected time points are shown in Fig. 2b. The mutation frequency of cells treated in mid-S phase was 0.15-0.22%, whereas the level in quiescent cells was 0.02-0.03%.
Psoralen Mutagenesis during the Cell Cycle-Our previous work (14) shows that the appearance of mutant cells required photoactivation. Thus, the actual mutagen was the psoralen cross-link. Consequently, it was of interest to question whether Hprt mutagenesis by free psoralen would show cell cycle variability. We treated cells at different phases of the cell cycle with psoralen (see "Materials and Methods") and processed them as before. Thioguanine-resistant colonies were recovered at frequencies a few fold over the background (0.004 -0.005 versus 0.001%), consistent with similar experiments with psoralen by other groups (37,38). However, in contrast to the previous experiment, the mutation frequency was essentially unchanged across the cell cycle (Fig. 2c).
Electroporation Controls and Specificity-Several control experiments were performed. Cells at different times of the cycle were treated with nonspecific TFOs or the specific TFO without photoactivation, or they were mock-electroporated and UVAtreated. We saw no increase in mutation frequency relative to cells that received no treatment in any of these experiments. We also measured the efficiency of electroporation in G 0 /G 1 and S phase cells using a green fluorescent protein reporter plasmid assay. We found only a modest difference in efficiency in S phase (60 -70% positive cells) as compared with G 0 /G 1 cells (40 -50%). The assay was repeated with fluorescent oligonucleotides. Approximately, 50% G 0 /G 1 cells showed nuclear fluorescence versus 80% for those in the S phase, again too small to explain the differences in Fig. 2b.
The Aprt gene has polypurine:polypyrimidine elements similar but not identical to the Exon 5 target in Hprt. There are straightforward selection protocols for identifying cells with mutations in Aprt. We have used this assay to monitor the specificity of the TFOs designed for the Hprt target. In previous work (14), we found that although AE-07 treatment mutagenized Hprt, it had no effect on Aprt. We repeated this experiment with cells in mid-S phase but again found no increase over background in the frequency of cells with mutations in the Aprt gene (data not shown).
The results of these experiments suggested that the differences in pso-TFO activity in quiescent and S phase cells were not artifacts of electroporation efficiency differences or because of a loss of target specificity (with an attendant increase in random mutagenesis) in S phase. The indifference to cell cycle position of mutagenesis by free psoralen indicated that there was a fundamental difference between the two reagents. Two nonexclusive explanations for the variability of the TFO-targeted mutagenesis appeared possible. The first explanation that the frequency of TFO targeted cross-links was the same in G 0 /G 1 and S phases and that the kinds of mutations or perhaps the efficiency of mutagenesis of the TFO cross-links varied.
Alternatively, the frequency of TFO-mediated cross-link formation might be different in G 0 /G 1 and S phases.
Kinds and Frequency of Mutations Induced by pso-TFO-The published spectra of psoralen-induced mutations in Hprt are dominated by base substitutions (37,38). We have performed experiments with mutation reporter plasmids carrying TFO-psoralen cross-links in the Hprt target sequence embedded in the supF reporter gene. These mutation profiles were also dominated by base substitutions. Almost all were located at T of the 5Ј-TA cross-link site (data not shown) (for schematic of the reporter gene, see Ref. 17). Targeted point mutations were also the major event in earlier studies (39,40) with psoralen-linked TFOs and supF reporter genes. However, in our analysis of pso-TFO-targeted mutations at the endogenous chromosomal Hprt target site, we found that ϳ90% thioguanine-resistant clones carried deletions in the target region extending into Exon 5 (14,17) and no point mutations were observed at the T of the 5Ј-TA cross-link site. This disparity could result from marked differences in the processing of the pso-TFO cross-link in the chromosomal target as compared with the shuttle vector plasmid (and the free psoralen) or might simply reflect the failure of the thioguanine resistance assay to report base substitutions at this site.
We tested the latter possibility by isolating DNA from colonies chosen at random following pso-TFO/UVA treatment (no selection was applied). The PCR products of the I4E5 target region from these clones were digested with XbaI whose recog- nition site is coincident with the cross-link target site (Fig. 1a). We screened 800 colonies from S phase experiments and recovered 54 digestion-resistant clones (6.8%) (Fig. 3). The analysis was performed on G 0 /G 1 colonies, and 6 of 731 were resistant (0.8%). Sequence analysis of resistant PCR fragments (43 S phase and 4 G 0 /G 1 phase) indicated that 41 of 43 were T 3 C at the T of the 5Ј-TA cross-link site (the others were point mutations at adjacent bases within the XbaI site). Two hundred colonies from mock-transfected/UVA-treated cells were also analyzed. There were no XbaI-resistant clones. These results showed that the actual frequency of targeted mutations was much higher (30-fold) than reported by the thioguanine selection. However, they confirmed the disparity between the G 0 /G 1 and S phase mutation frequencies.
Pso-TFO-mediated Cross-linking in Synchronized Cells-The difference in targeted mutation frequency between quiescent and S phase cells could be the result of cell cycle variability in mutagenesis functions (41) or a difference in the efficiency of pso-TFO cross-link formation. To test the latter possibility, we treated G 0 /G 1 and S phase cells with the pso-TFO and UVA and then isolated total genomic DNA. We used the classical denaturation resistance technique to measure the level of crosslinks in a restriction fragment containing the target sequence (42). The DNA was digested with EcoRI, and the digests were denatured and electrophoresed on neutral agarose gels. The gels were blotted onto a nylon filter and hybridized with a probe specific for the 2-kb fragment containing the target sequence. In initial control experiments, purified genomic DNA was incubated with the pso-TFO and was UVA-treated in vitro, restricted, and then either run separately or mixed with an equal amount of untreated DNA. The hybridization pattern showed clear resolution between the cross-linked and noncross-linked samples and run separately or in mixture (Fig.  4a). The hybridization signal of the cross-linked fragment was ϳ50% of the non-cross-linked control, reflecting the reduced efficiency of hybridization and/or the reduced retention on the filter of cross-linked DNA. The analysis of the DNA from the treated cells showed a greater extent of cross-linking in S phase than in G 0 /G 1. Quantitation by PhosphorImager indicated that ϳ19 Ϯ 2% S phase DNA was cross-linked (seven experiments), whereas the G 0 level was 4 Ϯ 2%. To control for nonspecific cross-linking, the blots were stripped and rehybridized with a probe to a 3-kb fragment of the Dhfr gene (Fig. 4b). There were no denaturation-resistant fragments in any of the samples, suggesting that the resistance was specific to the fragment containing the target sequence.
The difference in targeted TFO-psoralen adducts in cells in S phase or G 0 /G 1 was also shown by restriction digestion protection. DNA from treated cells was digested with EcoRI and XbaI followed by blotting and hybridization. The level of the XbaIresistant fragment was severalfold higher in the S phase (25-30% protection) sample than G 0 /G 1 (4 -5%) (Fig. 4c). The greater signal from this assay may reflect the inhibition of the restriction enzyme by both psoralen mono-adducts and crosslinks, whereas only the cross-links are reported by the denaturation resistance assay.

DISCUSSION
The pso-TFOs (43) are two component reagents with the extent of cross-linking (and the mutational consequences thereof) dependent upon and serving as a measure of the bound TFO at the time of photoactivation. The data presented here and in our previous publications (17,18) show that both the chemistry of the oligonucleotide and the biology of the cell are important determinants of TFO binding at a chromosomal target.
The biochemical measurement of the frequency of targeted cross-links combined with the frequency and kinds of mutations affords an initial insight into the relative importance of the several pathways available for processing cross-links. Approximately, 20 -30% of the target sequences were cross-linked when the cells were treated in S phase. However, the overall mutation frequency was ϳ7%. This indicates that as many as 75% of the cross-links were resolved without mutagenesis. Ͼ95% of the mutations were single base substitutions, whereas the deletions accounted for the remainder. Mutagenesis and repair of psoralen cross-links have been examined by many groups, and multiple pathways are involved (44 -48). Current models involve the generation of single strand gaps at the cross-link site, perhaps requiring the activity of the XPF⅐ERCC1 (XPF, the gene mutated in Xeroderma pigmentosum, complementation group F; ERCC1, excision repair cross complementing gene 1) complex (49). Recombinational repair of the gap using another intact copy of the target region as a donor would yield an unaltered version of the original target. Alternatively, the gap could serve as a template for synthesis by one or another of the damage bypass polymerases (50). This synthesis could have mutagenic consequences or might be error-free. Presumably, this would be the source of the point mutations. A second cycle of excision repair and synthesis would eventually produce either a mutant or wild type target sequence. Our base substitution data are in good agreement with a previous study (51) of pso-TFO mutagenesis of plasmids in yeast that elucidated mutagenesis from the furan or pyrone side of the psoralen cross-link (51,52). This work showed mutation patterns consistent with preferential excision of the duplex on the transcribed strand and also on the furan side of the cross-link. When both conditions applied as would be the case with the I4E5 target, the mutations were generated at the T at the pyrone side of the cross-link.
The majority of thioguanine-resistant clones contains deletions in which the psoralen target site and flanking sequence are lost (occasionally, we recover mutations at the A of the TA cross-link site). The deletions extend at least through the splice acceptor site, the minimum required to inactivate the gene. Our earlier sequence analysis of the deletions suggested that most of these were derived from double strand breaks (14,17). Deletions could be caused by successive cycles of incisions by excision repair enzymes or as a consequence of collisions with replication forks (53). Whatever the molecular pathways for the deletions, they are exercised much less frequently than those producing point mutations.
The recombinational repair pathway would not be operable in quiescent cells but would be available during the mid-S phase of our experiments as Hprt replicates early in S phase (54,55). The similar ratios of base substitution and deletion mutations (albeit at different frequencies) from G 0 /G 1 and S phase cells may indicate that the same mutation pathways operate in both conditions. Alternatively, it is also possible that the mutant cells recovered from the G 0 /G 1 cells were actually derived from a small number of S phase cells that escaped quiescence.
Why the greater frequency of targeting in S phase? There would seem to be at least three possible explanations. One is that the G 0 /G 1 cells are inefficiently electroporated relative to S phase cells. Our analysis of the electroporation efficiencies of the green fluorescent protein plasmid and the fluorescent oligonucleotides does not appear to support this contention. It is true that the physical presence of the fluorescent oligonucleotide in the nuclei of the G 0 /G 1 cells is not evidence of a functional presence. However, the green fluorescent protein plasmid results clearly reflect a functional, nuclear presence, and our experience has been that plasmid and oligonucleotide electroporation efficiencies are well correlated. A second possibility is that the stability of triplexes once formed is much greater in S phase cells than G 0 /G 1 cells. We have shown that triplexes are less stable in cells than in "physiological buffer" in vitro, perhaps because of cellular enzymes that can disrupt triplexes (56). These include helicases and the translocases found in chromatin-remodeling complexes (57)(58)(59). However, this suggestion is somewhat counterintuitive since we generally associate chromatin remodeling with gene activation and Hprt gene expression is quite low in quiescent cells (60).
A third possibility is that the levels of TFO binding reflect the accessibility of the target sequence. Triplex formation by 15-20 nucleotide TFOs on nucleosomal sequences is impeded by the requirement of the third strand to occupy the major groove, 8 -10 base pairs of which would be exposed with the remainder turned against the histone core complex. In vitro studies indicate that nucleosomal sequences associated with the nucleosomal core are inaccessible to TFOs, whereas those at the ends or in internucleosomal linker regions are more available (20 -22, 61). Although direct experiments have not been performed, one implication is that nucleosomal DNA whose association with the histone core has been relaxed by ATP-dependent remodeling (62)(63)(64) or certain histone modifications (65) would be more open for triplex formation. The Hprt gene is a member of a group of constitutively expressed housekeeping genes (including DHFR) whose expression is quite low in quiescent cells (60,66). Transcript levels increase markedly upon growth stimulation and, in the case of DHFR, are dependent on E2F transcription factors (67). This family also includes repressors of E2F-dependent promoters, which are involved in silencing genes in quiescent cells via mechanisms that include alterations in histone modification and chromatin structure (68,69). Assuming that these alterations extend to nucleosomes downstream of the promoter region, it may be that TFO bioactivity reflects the chromatin structural dynamics of the Hprt target.
The results presented here have practical significance. TFOs have been of interest for many years because of potential applications as gene-targeting reagents for manipulation of genomic sequences. The Hprt gene target has been instrumental in the development of bioactive TFOs because of the opportunity for facile and quantitative selection of cells with targeted events. However, most genes of interest do not have a selection option available for identification and analysis of cells with the desired end point. Thus, it will be important to develop protocols that permit recovery of clones of interest with a minimum of effort. The data presented here suggest that successful pro- FIG. 4. a, denaturation resistance assay of targeted cross-links. Genomic DNA treated in vitro or in vivo with AE-07, was digested with EcoRI, which releases a 2-kb fragment containing the target site. Samples with the exception of lane 4 were denatured and electrophoresed, blotted, and hybridized with a probe to the 2-kb EcoRI fragment. Lane 1, genomic DNA cross-linked in vitro by AE-07; lane 2, non-cross-linked DNA; lane 3, mixture of equal amounts of lanes 1 and 2; lane 4, non-cross-linked, non-denatured DNA. Denatured DNA from untreated control cells (lane C) or AE-07-treated G 0 /G 1 or S phase cells (two independent experiments). The arrowhead marks the position of the denatured fragment, and the arrow marks the non-denatured or denaturation-resistant fragment. b, hybridization with a probe to a 3-kb (arrow) fragment of the DHFR gene. Samples were not denatured (C ND ) or denatured from control (C) or AE-07-treated G 0 /G 1 or S phase cells. c, blot of EcoRI and XbaI digestion of samples from AE-07-treated cells. Control DNA digested with EcoRI (C R ) or EcoRI and XbaI (S RX or C RX ). DNA from AE-07/UVA G 0 /G 1 or S phase cells digested with EcoRI and XbaI. The arrow marks the position of the XbaI-resistant fragment. cedures will require effective oligonucleotide chemistry combined with manipulation of the cell biology. This will be an important consideration for gene targeting in hematopoietic stem cells, which are largely quiescent (70).