Altered repair of targeted psoralen photoadducts in the context of an oligonucleotide-mediated triple helix.

Oligonucleotides can bind as third strands of DNA in a sequence-specific manner to form triple helices. Psoralen-conjugated, triplex-forming oligonucleotides (TFOs) have been used for the site-specific modification of DNA to inhibit transcription and to target mutations to selected genes. Such strategies, however, must take into account the ability of the cell to repair the triplex-directed lesion. We report experiments showing that the pattern of mutations produced by triplex-targeted psoralen adducts in an SV40 shuttle vector in monkey COS cells can be influenced by the associated third strand. Mutations induced by psoralen adducts in the context of a TFO of length 10 were the same as those generated by isolated adducts but were found to be different from those generated in the presence of a TFO of length 30 at the same target site. In complementary experiments, HeLa whole cell extracts were used to directly assess repair of the TFO-directed psoralen adducts in vitro. Excision of the damaged DNA was inhibited in the context of the 30-mer TFO, but not the 10-mer. These results suggest that an extended triple helix of length 30, which exceeds the typical size of the nucleotide excision repair patch in mammalian cells, can alter repair of an associated psoralen adduct. We present a model correlating these results and proposing that the incision steps in nucleotide excision repair in mammalian cells can be blocked by the presence of a third strand of sufficient length and binding affinity, thereby changing the pattern of mutations. These results may have implications for the use of triplex-forming oligonucleotides for genetic manipulation, and they may lead to the use of such oligonucleotides as tools to probe DNA repair pathways.

Oligonucleotides can bind in the major groove of duplex DNA to form triple helices in a sequence-specific manner (1)(2)(3)(4). Progress in elucidating the third strand binding code has raised the possibility of developing nucleic acids as sequencespecific reagents for research and possibly therapeutic applications. Oligonucleotide-mediated triplex formation has been shown to inhibit transcription factor binding to promoter sites and to block transcription in vitro and in vivo (5). Strategies to enhance transcription inhibition using oligonucleotide-intercalator conjugates have also been employed (6,7). Triplex formation has further been used as a tool to generate unique cleavage sites in DNA in vitro (8). We and others have explored the use of triplex-forming oligonucleotides (TFOs) 1 as a method to deliver a tethered mutagen to a selected gene for the site-specific introduction of DNA damage (9 -12). Psoralen-conjugated oligonucleotides directed to a site in a mutation reporter gene were found to induce targeted mutations in (10) and SV40 (9) DNA in experiments in which the triplex formation was carried out in vitro. In recent work, conditions were determined under which targeted mutagenesis of an SV40 vector could be achieved in vivo via intracellular triplex formation using psoralen-linked TFOs (13), raising the possibility of developing oligonucleotides as sequence-specific reagents for in vivo genetic manipulation.
In using a TFO to deliver a tethered mutagen to a target site, however, the effect of the oligonucleotide on the cellular processes of repair and replication must be considered. Degols et al. (14) have shown using a polymerase chain reaction-based assay that psoralen adducts targeted by a 15-nucleotide TFO are removed from a plasmid transfected into human cells, but they did not examine how the third strand might influence this process. In using a series of psoralen-linked TFOs to target mutations to an SV40 vector both in vitro (9) and in vivo (13), we observed in preliminary work some variability in the spectra of targeted mutations in the context of the triple helix at the same target site. Because triple helix formation has been shown to inhibit restriction enzymes (15)(16)(17) and to resist deoxyribonuclease I digestion (18), we hypothesized that it might also interfere with repair activities and thereby affect mutagenesis by the targeted adducts.
In mammalian cells, psoralen adducts and other types of DNA damage are repaired by the nucleotide excision repair (NER) pathway (19,20), in which a multiprotein complex recognizes and excises the damage via dual incisions in the damaged strand 6 nucleotides 3Ј and 22 nucleotides 5Ј from the site of damage (21). Release of the damaged DNA fragment via helicase activity is followed by gap-filling repair synthesis (19). Based on this model of NER, we have designed experiments to examine the effect of a triple helix on DNA repair and mutagenesis. We have used two TFOs of different length (either shorter or longer than the size of the usual NER repair patch) to direct specific psoralen adducts to the same target site in a reporter gene. We have also developed a strategy to deliver targeted psoralen adducts using an oligonucleotide that can be subsequently detached from the adduct by disulfide bond reduction. In this way, psoralen mutagenesis at the same site of adduct formation can be directly compared in the presence of triple helices of differing lengths as well as in the absence of an associated third strand.
We report here that, in shuttle vector mutagenesis experiments in monkey COS cells, psoralen-linked TFOs of length 10 versus 30 generate different mutational spectra at the same target site, with mutagenesis in the context of the 10-mer being similar to that produced by a psoralen adduct alone. In these experiments, we also find a difference between the spectrum of mutations generated by psoralen monoadducts and cross-links targeted to the same intercalation site, a result which has additional implications for models of interstrand cross-link repair. Furthermore, using human cell extracts to study the repair of the shuttle vector-psoralen-TFO substrate in vitro, we show that a triple helix of sufficient length and binding affinity can inhibit excision of the damaged DNA. We propose a model correlating these results and suggesting that the incision steps in NER can be blocked by the presence of a third strand.

MATERIALS AND METHODS
Oligonucleotides and Vectors-Psoralen-linked oligonucleotides were obtained from Oligos Etc. (Wilsonville, OR) or were synthesized by J. Flory of the W. M. Keck Biotechnology Resource Center at Yale using materials from Glen Research (Sterling, VA). The psoralen is incorporated into the oligonucleotide synthesis as a psoralen phosphoramidite, resulting in oligonucleotides linked at their 5Ј end via a two-carbon linker arm to 4Ј-hydroxymethyl-4,5Ј,8-trimethylpsoralen, either pso-AG10 or pso-AG30 (Fig. 1). Pso-S-S-AG10, containing a disulfide bond in the linker arm connecting the psoralen to the oligonucleotide, was synthesized by incorporation of a thiol modifier (S-S) phosphoramidite (Glen Research) into the automated synthesis just prior to the addition of the 5Ј-psoralen, yielding an oligonucleotide with the same sequence as pso-AG10, but with the psoralen and the oligonucleotide separable by disulfide bond reduction. To preserve the disulfide linker in the synthesis, modification of the oxidation step was required, as directed by Glen Research.
The SV40 shuttle vector, pSupFG1, was previously constructed from the vector pSP189 to contain a modified supF amber suppressor tRNA gene (supFG1) as a mutation reporter gene incorporating a 30-base pair polypurine/polypyrimidine site amenable to triple helix formation (13).
Triplex Formation and Targeted Photoadduct Formation-The SV40 vector DNA at 80 nM was incubated with either psoralen-AG10 or psoralen-AG30 at 1 M and irradiated with either long wavelength UV light (UVA, 320 -400 nm) at a dose of 1.8 J/cm 2 or with visible light (centered at 447 nm) at a dose of 11.7 J/cm 2 . Irradiation was delivered using 365 and 447 lamps obtained from Southern New England Ultraviolet, Inc. (Branford, CT). A window glass filter was used to eliminate UVB radiation. A radiometer equipped with filters for UVA and visible light (obtained from International Light, Newburyport, MA) was used to measure the lamp output (typical UVA irradiance of 5-7 milliwatts/ cm 2 at 320 -400 nm). The formation of specific photoadducts (monoadducts and cross-links) was monitored by gel mobility shift assays and confirmed by HPLC analysis (12).
In the case of pso-S-S-AG10, the irradiated mixture was further incubated with 50 mM dithiothreitol at 55°C for 3-4 h to release the oligonucleotide by disulfide bond reduction. The detached oligonucleotide was separated from the duplex by heating the mixture at 65°C for 5 min and removed by filtration through a Centricon-100 filter (Amicon, Beverly, MA), yielding pSupFG1 containing a site-specific psoralen adduct without an associated oligonucleotide.
Successful detachment of the oligonucleotide from the targeted psoralen adduct was monitored by end-labeling pso-S-S-AG10 with [␣-32 P]dideoxy-ATP (Amersham Corp.) and terminal transferase. The sample was analyzed before and after dithiothreitol treatment and Centricon-100 filtration by denaturing gel electrophoresis and autoradiography.
Shuttle Vector Mutagenesis-The plasmid-psoralen-oligonucleotide complexes were transfected into monkey COS-7 cells (ATCC 1651-CRL) via electroporation as described previously (13). Following 48 h to allow repair and replication, the SV40 vector DNA was harvested from the COS cells using a modified alkaline lysis procedure, as described (13). The rescued vector DNA was used to transform either Escherichia coli SY204 (lacZ125(amber)) (22) or MBM7070 (lacZ(amber)) (23) by electroporation (Bio-Rad, setting 25 microfarads/250 watts/1800 V, using a 0.1-cm cuvette). The transformed E. coli cells were plated onto LB plates containing 50 g/ml of ampicillin, 100 g/ml of 5-bromo-4-chloro-3-indoyl ␤-D-galactoside, and 1 M isopropyl-␤-D-thiogalactopyranoside and were incubated at 37°C overnight. Mutant colonies containing inactivated supFG1 genes unable to suppress the amber mutation in the host cell ␤-galactosidase gene were detected as white colonies among the wild type blue ones. The mutant colonies and the total colonies were counted. The mutant colonies were purified, and the plasmids were isolated for DNA sequence analysis (13).
In Vitro Repair Assays-HeLa cell-free extracts were prepared as described previously (24). Oligonucleotides pso-AG10 and pso-AG30 were 3Ј end-labeled using terminal transferase and [␣-32 P]dideoxy-ATP (Amersham). Triplex formation with the target plasmid DNA (pSupFG1) was performed as described above. Following triplex formation and photoadduct generation, the majority of the unbound oligonucleotide was removed from the sample by dilution and filtration using Centricon-100 filters (Amicon, Beverly, MA).
Repair reactions were carried out essentially as described (25), with slight modifications. Plasmid-psoralen-oligonucleotide DNAs at 2 ϫ 10 Ϫ9 M were incubated for 3 h at 30°C in the HeLa cell extracts containing 10 -15 g/l protein and supplemented as described (25). The DNA was extracted with phenol/choloroform and concentrated by ethanol precipitation. The samples were diluted 10-fold with formamide, boiled for 5 min, and analyzed by polyacrylamide gel electrophoresis in an 8% gel containing 7 M urea, followed by autoradiography.

RESULTS
The experimental protocol used to study mutagenesis in mammalian cells induced by triple helix-directed psoralen photoadducts is diagrammed in Fig. 1. The assay is based on the use of an SV40-based shuttle vector, pSupFG1 (13). This vector contains the supFG1 gene, an amber suppressor tyrosine tRNA gene of E. coli, as a mutation reporter gene, along with both the SV40 and pBR327 origins of replication. The supFG1 gene is a functional derivative of the supF gene that has been modified by the incorporation of a 30-base pair polypurine/polypyrimidine site for triple helix formation at its 3Ј end (13). Psoralenconjugated oligonucleotides pso-AG10 and pso-AG30 were designed to bind as third strands to base pairs 167-176 and base pairs 167-196 of the supFG1 gene, respectively, in the antiparallel triple helix motif (3,26). Triplex formation by either oligonucleotide positions the psoralen conjugate for intercalation between base pairs 166 and 167. In previous work, we have shown that pso-AG10 and pso-AG30 bind to the target site in FIG. 1. Experimental strategy to study mutagenesis mediated by triple helix-directed psoralen adducts. The triplex-forming oligonucleotides, either psoralen-AG10 or psoralen-AG30, are shown directly above their targeted sequences in the supFG1 gene (base pairs 167-176 and base pairs 167-196, respectively), contained within the SV40 vector, pSupFG1. The structure of the tethered psoralen (4Јhydroxymethyl-4,5Ј-8-trimethylpsoralen attached at the 4Ј-hydroxymethyl position via a two-carbon linker arm to the 5Ј-phosphate of the oligonucleotide) is also shown. The psoralen-oligonucleotides are incubated with the SV40 vector DNA to allow site-specific triplex formation. Photoactivation of the psoralen by irradiation with either long wave ultraviolet light (UVA, 365 nm) or visible light (447 nm) is designed to generate adducts at the targeted intercalation site (166 -167), as indicated by the arrow. The oligomer-plasmid complex is then transfected into monkey COS-7 cells and allowed to replicate for 48 h. Following purification of the vector DNA by an alkaline lysis procedure, the DNA is used to transform E. coli SY204 lacZ125 (Am). Transformants are selected on ampicillin plates containing 5-bromo-4-chloro-3-indoyl ␤-Dgalactoside and isopropyl-␤-D-thiogalactopyranoside for detection and isolation of mutants (white colonies) in which the supFG1 gene has been inactivated by mutation. the supFG1 gene with equilibrium dissociation constants (K d ) of 8 ϫ 10 Ϫ7 M and 3 ϫ 10 Ϫ9 M, respectively (13).
The TFO and the vector DNA are coincubated in vitro to allow triplex formation, and either long wave UV light (UVA, centered at 365 nm) or visible light (447 nm) is used to photoactivate the psoralen to generate targeted photoadducts at base pairs 166 -167 of the supFG1 gene. At the UVA dose of 1.8 J/cm 2 used in the present experiments, a previous analysis of the photoproducts in the context of the triple helix (using gel mobility shift assays and HPLC analysis) revealed that approximately 65% of the plasmid DNAs have a psoralen interstrand cross-link between the thymidines in base pairs 166 -167, with almost all of the cross-links oriented such that the furan-side adduct is on the T in base pair 166 and the pyrone-side adduct is on the T in base pair 167. In addition, 20% have a psoralen furan-side monoadduct at the T in base pair 166, 10% have a monoadduct (predominately pyrone-side) at the T in base pair 167, and 5% are not covalently modified (12). At a visible light dose of 11.7 J/cm 2 , only 1% of the vectors have a psoralen interstrand cross-link between the Ts at 166 -167, whereas 41% have a furan-side monoadduct on the T in base pair 166, 1% have a monoadduct on the T in base pair 167, and approximately 58% of the plasmids are not covalently modified (12). A monoadduct in the present discussion refers to the photoaddition of the tethered psoralen to just one strand of the duplex target, yielding a structure in which the TFO is covalently linked to that strand. A cross-link means that the psoralen has photoreacted with both strands of the duplex, yielding a structure in which the TFO is thereby covalently linked to both strands of the target DNA. A furan-side adduct refers to a lesion formed by cycloaddition to thymidine at the 4Ј,5Ј bond of the furan ring of the psoralen, whereas a pyrone-side adduct refers to the product of cycloaddition involving the 3,4 bond of the pyrone ring. Although the previously published work focuses on photoadduct formation directed by pso-AG10, similar proportions of adduct formation have been observed with pso-AG30 (data not shown).
The plasmid-psoralen-oligonucleotide complex is transfected into monkey COS cells by electroporation. After 48 h for repair and/or replication, the vector DNA is isolated and used to transform lacZ (amber) bacteria for the genetic analysis of the supF gene. Prior to the bacterial transformation, the DNA is subjected to digestion with DpnI, which restricts nonreplicated vector DNA that has not acquired the mammalian methylation pattern. This step eliminates unprocessed input molecules that might lead to misleading results. Table I presents a comparison of the induced mutation frequencies in pSupFG1 following treatment with either pso-AG10 or pso-AG30 and either UVA or visible light. For both TFOs, UVA photoactivation of the tethered psoralen leads to more mutations than does visible light activation. This is not surprising, since the overall level of photoadduct formation is much higher with UVA. However, pso-AG10 appears to more efficiently generate mutations than does pso-AG30. This difference provided a preliminary indication that the third strand can affect the repair and mutagenesis of the damaged vector.
The mutants induced by pso-AG10 and pso-AG30 were analyzed by DNA sequencing (Fig. 2). With visible light activation, both pso-AG10 and pso-AG30 produced predominately T:A to A:T transversions at base pair 166. Since the main photoadduct following visible light irradiation is a furan-side psoralen monoadduct on the T in that base pair, these results are consistent with mutagenesis in both cases arising from translesion DNA synthesis across this adduct.
However, the mutations induced following UVA irradiation are distinct from those after visible light. Treatment of the vector with pso-AG10 and UVA generated mostly A:T to T:A transversions at base pair 167. This shift in mutation position correlates with the change in photoadduct distribution following UVA as opposed to visible light. After UVA, 75% of the plasmids have an adduct involving the T at 167, either a monoadduct (10%) or a cross-link (65%). Although the targeted cross-links consist of lesions at both 166 and 167, the data indicate that cross-link repair in this case leads predominately to mutations at 167.
In contrast, in the case of pso-AG30 plus UVA, A:T to T:A transversions are not seen at position 167. Some are found at 166, but the overall spectrum is, in general, more diverse, with mutations occurring over several base pairs around the psoralen intercalation site, including base pairs 165, 166, 167, and 168. In addition, although deletions around the target site are seen with both pso-AG10 and pso-AG30, larger deletions were seen with pso-AG30. These clear differences in the mutation spectra induced by pso-AG10 versus pso-AG30 suggest that repair of a triplex-targeted psoralen adduct can be influenced by the associated third strand.
Because the 30-mer binds more tightly to the target duplex than the 10-mer, and because it forms a triple helix that is longer than the standard nucleotide excision repair patch, we hypothesized that the pattern of mutations it induced was likely to reflect the effect of a third strand on repair to a much larger extent than in the case of the 10-mer. However, it remained a possibility that mutagenesis in the context of the 10-mer was also abnormal. Therefore, we designed an oligonucleotide that could deliver the psoralen to the target site at base pairs 166 -167, but which could be detached from the adduct once it had been formed. To do this, an oligomer, pso-S-S-AG10, was synthesized to contain a disulfide bond in the linker between the psoralen and the 5Ј A in the AG10 sequence. Using this oligomer in conjunction with UVA photoactivation, targeted psoralen adducts were produced in pSupFG1. Release of the oligonucleotide fragment was achieved by treatment with dithiothreitol and heat, followed by size filtration to remove it from the sample and so prevent reformation of the triple helix. Using pso-S-S-AG10 3Ј end-labeled with 32 P, successful detachment of the otherwise covalently attached oligomer was confirmed by gel electrophoresis and autoradiography of the vector DNA before and after oligonucleotide release (data not shown; complete details of this technique will be published elsewhere).
The vector DNA containing the site-specific adducts (but not the targeting oligonucleotide) was used to transfect COS cells as above, and the resulting mutations were analyzed. Out of 14 point mutations, 12 were found at 167, similar to those generated in the context of the 10-mer in the case of UVA photoactivation. Ten small deletions were also seen encompassing the target site. These results suggest that isolated psoralen adducts and adducts in the context of the 10-mer are handled by the cells in a qualitatively similar fashion. In contrast, pso- AG30 adducts are processed in an altered way.
In order to further elucidate potential differences in repair in the context of a triple helix, we examined repair of the plasmidpsoralen-oligonucleotide substrate in vitro in human cell extracts. We designed an assay to ask whether the triplex-targeted psoralen adduct is excised as predicted by the NER model of DNA damage repair. It has been shown in human cell extracts that dual incisions are made in the damaged DNA strand 6 nucleotides on the 3Ј side and 22 nucleotides on the 5Ј side of the damage (21). Helicase activity removes the damaged oligonucleotide fragment, and polymerase activity fills in the resulting gap (27). Based on this model, we hypothesized that excision repair of a triplex-targeted monoadduct would yield an oligonucleotide fragment containing a psoralen adduct tethered to the TFO, constituting a branched structure. Repair of a cross-link is envisioned to be more complicated, requiring sequential damage excision and repair synthesis on the two strands and/or interstrand recombination, as has been proposed (28 -32). Hence, the expected products that may result from in vitro repair of a cross-linked substrate are less predictable.
To detect potential repair products, the TFOs were radioactively labeled at the 3Ј end using [␣-32 P]dideoxy-ATP and terminal transferase. The labeled TFOs, in conjunction with either UVA or visible light, were used to generate triplextargeted psoralen adducts on pSupFG1.
The psoralen-oligonucleotide-damaged plasmids were incubated in HeLa cell extracts under conditions as described for NER assays (25), and the products were analyzed by denaturing gel polyacrylamide electrophoresis and autoradiography (Figs. 3 and 4). Note that under the conditions of this gel analysis, neither triplex nor duplex interactions will persist; all noncovalent interactions will be disrupted, and the DNA fragments will be denatured. In Fig. 3, lanes 1 and 2 show as controls the input substrates formed by pso-AG10 and pso-AG30, respectively, incubated with pSupFG1 and irradiated with visible light. These samples include the plasmid covalently attached to the labeled psoralen-TFO (which is too large to enter the gel and is stuck at the top) and a residual amount of the labeled but unbound pso-TFO (which was not completely removed in the filtration step during sample preparation). Bands corresponding to pso-AG10 and pso-AG30 are therefore visible in these lanes (indicated by the arrows).
When the substrate from lane 1 (a pso-AG10-directed monoadduct) is added to the HeLa extract (lane 3), the extract generates a species of reduced mobility compared with the labeled pso-AG10 itself, consistent with a branched structure arising from excision from the plasmid of a damaged single strand fragment containing the psoralen-AG10 adduct. Much of the unbound, input pso-AG10 is degraded in the reactions and so is not visualized. In contrast to the case with pso-AG10, when a pso-AG30 and visible light-damaged plasmid is used as a substrate in the repair reaction (lane 4), no species of reduced mobility relative to pso-AG30 that might represent fragments released in repair are visualized.
When the substrate in the HeLa extract consists of pSupFG1 treated with pso-AG10 and UVA (Fig. 4, lane 2), a labeled fragment of reduced mobility relative to pso-AG10 alone (compare lane 1) is released from the plasmid. This band represents a fragment similar to that seen in Fig. 3, lane 3, in which the substrate was photoactivated by visible light. Since UVA treatment of pso-AG10/pSupFG1 generates both monoadducts and cross-links, whereas visible light produces only monoadducts, this species therefore likely arises from repair of the pso-AG10 monoadducts in the UVA-irradiated sample.
In the case of pso-AG30 and UVA-induced damage (Fig. 4,  lanes 3 and 4), excision of a damaged DNA fragment appears to be inhibited. The position of the free pso-AG30 oligomer is shown in lane 4, and the results of incubation of the pso-AG30 and UVA-damaged plasmid in the HeLa extracts are shown in  No discrete band that might correspond to an excised damaged fragment is seen. Taken together, these results suggest that repair of the pso-AG30-targeted psoralen adducts is hindered by the long, tightly bound third strand. DISCUSSION We have examined psoralen mutagenesis targeted by triplex-forming oligonucleotides in the supFG1 gene carried in an SV40 shuttle vector passaged in COS cells. Although both pso-AG10 and pso-AG30 generated targeted mutations in the supFG1 following in vitro triplex formation and photoactivation, several differences were observed. Pso-AG10 generated higher mutation frequencies than did pso-AG30, especially following UVA irradiation. Also, the spectra of induced mutations following UVA irradiation differed between pso-AG10 and pso-AG30, with pso-AG10 generating mutations similar to those seen with isolated psoralen adducts at the same site. These results suggest that an associated triple helix can influence the repair and processing of targeted psoralen adducts, depending on the length and/or binding affinity of the third strand. The in vitro repair experiments provide further evidence for the effect of the TFO on repair. These revealed an apparent inhibition of excision repair in the context of the 30-mer TFO in comparison with the 10-mer, as measured by release of a DNA fragment containing the psoralen-oligonucleotide adduct.
The results reported here confirm and extend our preliminary observations, suggesting differences in mutagenesis by targeted psoralen adducts depending on the associated TFO (9,13). In work in which the triplex-targeted adduct formation was carried out in vitro using pso-AG10 and UVA (9), we had previously observed mutations mostly at base pair 167. However, in experiments in which pso-AG30 was used to treat cells in culture (already containing the shuttle vector) and thereby to target psoralen adducts via in vivo triplex formation, we found mutations mostly at base pair 166 (13). Initially, we were concerned that there might be some difference in photochemistry inside versus outside of the cells that might alter the mutation distribution. However, the present study provides the explanation that the associated TFO, itself, can influence psoralen adduct repair and mutagenesis.
A model to explain these results is illustrated in Fig. 5. In this diagram, using the psoralen adduct as a reference point, the potential endonuclease incisions in the NER pathway in each case are shown 6 nucleotides on the 3Ј side of the lesion and 22 nucleotides on the 5Ј side, as proposed for mammalian cells (21) (see Fig. 1 for the sequences and polarity of the strands in this diagram). Note that for both pso-AG10 and pso-AG30, the psoralen-TFO monoadducts occur mostly on the T in base pair 166 (the upper, purine-rich strand of the target duplex). We propose that excision of the pso-AG30 monoadducts (Fig. 5A, 3) is less efficient than of the pso-AG10 monoadducts (Fig. 5A, 1), because the longer triple helix overlaps the standard excision repair patch, potentially blocking the 5Ј incision. Also, the binding affinity of the 30-mer is 270-fold greater than that of the 10-mer, and so the 30-mer would be more likely to compete with and block NER proteins. The inhibition of repair endonuclease activity would be consistent with the inhibition of restriction enzymes by third strands that

FIG. 3. Excision repair of visible light-activated psoralen-oligonucleotide adducts in HeLa cell extracts.
Triplex-forming oligonucleotides, pso-AG10 and pso-AG30, were radiolabeled and incubated with the pSupFG1 vector to form site-specific triple helices. Visible light photoactivation was used to generate targeted monoadducts (12), consisting of the psoralen-oligonucleotides covalently linked to one strand of the target duplex. The damaged plasmids were incubated in HeLa whole cell extracts, which have been shown to contain DNA repair activity (24,34,35). The resulting products were analyzed by denaturing polyacrylamide gel electrophoresis and autoradiography. Lanes 1 (pso-AG10) and 3 (pso-AG30) show the initial samples before incubation in the extracts. These samples include some unreacted psoralen-oligomers, the positions of which are indicated by the arrows. Lanes 2 (pso-AG10) and 4 (pso-AG30) show the products of the incubation in the HeLa cell extracts.

FIG. 4. Excision repair of UVA-activated psoralen-oligonucleotide adducts in HeLa cell extracts.
Triplex-forming oligonucleotides, pso-AG10 and pso-AG30, were radiolabeled and incubated with the pSupFG1 vector to form site-specific triple helices. UVA photoactivation was used to generate a mixture of targeted monoadducts and cross-links (12), consisting of the psoralen-oligonucleotides covalently linked to either one strand of the target duplex or to both strands of the duplex, respectively. The damaged plasmids were incubated in HeLa whole cell extracts (24,34,35). The products were analyzed by denaturing polyacrylamide gel electrophoresis and autoradiography. Lanes 1 (pso-AG10) and 4 (pso-AG30) show the samples prior to incubation in the extracts. These samples include some unreacted psoralen-oligomers, the positions of which are indicated by the arrows. Lanes 2 (pso-AG10) and 3 (pso-AG30) show the products of the HeLa cell extract incubation.
overlap their recognition sites (15)(16)(17). Such an effect would account for the difference in the release of the damaged DNA fragments, as shown in Figs. 3 and 4.
Since monoadduct repair should be relatively error-free, mutagenesis by the pso-TFO monoadducts, both in the case of pso-AG10 and pso-AG30, likely arises during bypass replication, as diagrammed in Fig. 5B. This model predicts that the induced mutation spectra for pso-AG10 and pso-AG30 monoadducts should be similar. As shown in Fig. 2, monoadducts targeted by the two psoralen-TFOs generated in both cases mostly T:A to A:T transversions at base pair 166, the proposed site of lesion bypass.
In the case of the UVA-induced cross-links, bypass replication would be blocked by the interstrand cross-link, and so much of the mutagenesis likely occurs during repair. Repair of psoralen cross-links is proposed to involve sequential excision repair, possibly coupled to recombinational pathways (28 -32). In the case of free psoralen (without an associated TFO), the choice of the initial strand to be repaired is influenced by the orientation of the psoralen cross-link (furan-side lesion preferentially repaired first in some systems (32)) and by the sequence context (33). For the pso-AG10 adduct (Fig. 5A, 2), repair may be biased toward the purine-rich strand (upper strand of the duplex in the diagram) for two reasons. First, the furan-side adduct in the cross-link is predominately on this strand (12). Second, the incisions on this strand would not be blocked by the triple helix, whereas the 3Ј incision on the other strand (6 nucleotides from the psoralen lesion) would overlap with the triple helix region. Following excision repair of the damaged fragment from the upper strand, gap-filling repair synthesis would have to bypass the remaining lesion on the lower strand at position 167, leading to a mutation at that site.
In accordance with this model, most of the mutations induced by pso-AG10 cross-links were A:T to T:A transversions at base pair 167, not at 166 as in the case of the monoadducts. This model for cross-link repair and mutagenesis is further supported by the experiment in which targeted psoralen adducts were generated by pso-S-S-AG10 and UVA and then detached from the targeting oligonucleotide by disulfide bond reduction. The resulting vector substrate, containing predominately site-specific cross-links, yielded mostly A:T to T:A transversions at 167. This is consistent with cross-link repair involving excision of the damaged fragment from the upper strand, followed by gap-filling repair synthesis requiring bypass of the remaining lesion at 167.
An alternative pathway, in which the initial gap is filled via strand transfer from an undamaged template, is also possible and has been proposed (28 -32). This would be followed by excision repair of the lower strand. In this case, an undamaged template for gap-filling synthesis would then be provided by the transferred strand. This scenario would therefore not require trans-lesion synthesis, and so it would be less likely to lead to mutations. This is consistent with the report of Sladek et al. (28) that psoralen cross-link-induced mutagenesis is suppressed in bacteria capable of recombination. Although this recombinational pathway may occur in mammalian cells, the pattern of targeted mutagenesis we have observed suggests that at least some of the cross-linked vectors are repaired in a mutagenic pathway that requires gap-filling repair synthesis across a damaged template.
With regard to pso-AG30-targeted cross-links, however, the potential sites of both the 5Ј incision on the upper strand and the 3Ј incision on the lower strand overlap the triple helix (Fig.  5A, 4). However, incision in the upper strand may be more strongly inhibited than in the lower strand, since the upper strand is the purine-rich strand to which the TFO directly binds by reverse Hoogsteen bonding in the antiparallel triple helix motif (3). If the incision in the lower strand is therefore less inhibited by the third strand oligonucleotide, then the second pathway illustrated in Fig. 5A, 4, would be favored. This pathway is predicted to involve bypass synthesis across a psoralen-damaged thymidine at position 166, and so it would be expected to lead to mutations at 166. As shown in Fig. 2, pso-AG30 plus UVA does lead to a plurality of T:A to A:T transversions at 166.
It is also possible, however, that the 166 mutations may not necessarily be a consequence of pso-AG30-directed cross-links. Instead, these mutations may arise, in part, from the 20% of the UVA-induced adducts that are monoadducts at 166. In addition, many of the mutations induced by pso-AG30 and UVA irradiation were found to occur not just at 166 but also at several base pairs (165-168) around the target site, and these are not specifically explained by the model. The mechanism by which these are generated remains to be elucidated.
If the usual cross-link repair pathway is, in fact, partially inhibited in the context of pso-AG30, we would expect that some mutations that normally arise in this pathway would be underrepresented in our assay. As shown in Fig. 2, no 167 A:T  Fig. 1). The psoralen-conjugated oligonucleotides are represented by the smaller third strands in each diagram, being connected to the duplex by either one line (monoadduct) or two lines (cross-link). The small arrows mark predicted sites of endonuclease incisions based on the reported properties of the nucleotide excision repair complex in mammalian cells, which generates incisions 6 nucleotides 3Ј and 22 nucleotides 5Ј of a lesion (21). Displaced arrows (as in A3 and A4) are meant to suggest possible inhibition of the endonuclease activity by the third strand. DNA synthesis, either as a component of the repair reaction or in trans-lesion bypass replication, is represented by the dashed lines.
to T:A transversions were observed. Also, the absence of a particular class of mutations might be expected to affect the overall mutation frequency. As shown in Table I, pso-AG30 induced fewer mutations than did pso-AG10, in agreement with this hypothesis.
The proposed model, however, does not specifically explain the deletions, which are seen mostly in the case of the targeted cross-links. It is possible that, in the attempted repair of both strands of a cross-linked substrate, double strand breaks are produced. The resolution of these may occasionally lead to the observed deletions. Also, recombinational pathways may contribute to the deletion mutagenesis. At this point, however, except for the association with cross-links, the genesis of the deletions is still to be determined.
In this work, we have presented evidence that psoralen adduct repair and mutagenesis can be influenced by an associated triple helix. Both the frequency and the spectrum of induced mutations were seen to vary depending on the length (and the binding affinity) of the TFO used to direct the psoralen adducts to the target site. In vitro experiments using HeLa cell extracts suggest that excision of damaged DNA containing a psoralen-TFO adduct is at least partially blocked by the third strand. The exact mechanism by which NER is inhibited by the triple helix, however, remains to be elucidated. Also, although the model proposed here can account for mutagenesis by pso-AG10targeted monoadducts and cross-links and by pso-AG30-targeted monoadducts, it does not fully explain the mutation distribution induced by pso-AG30-directed cross-links. Further work in this regard is required to examine the possible pathways of cross-link repair and mutagenesis in the context of an extended triple helix. Nonetheless, the results presented here, suggesting that triple helices can interfere with normal DNA repair pathways, bear on the use of TFOs for genetic manipulation. Furthermore, since triple helices can at least partially inhibit NER, TFOs may have utility as tools to probe DNA repair pathways.