ABSTRACT
Triple helices represent an attractive method for modulating specific gene expression. In particular, cross-linking between a triplex-forming oligonucleotide (TFO) and its duplex DNA target, typically through the formation of psoralen photoadducts, allows efficient blocking of elongation by RNA polymerases in vitro. However, in vivo, this approach is limited by DNA repair of the photoadduct. Here we describe the use of an oligodeoxyribonucleotide 19mer psoralen-modified TFO to form covalent linkages between an oligonucleotide and both strands of the targeted duplex DNA, thereby efficiently blocking expression of a luciferase reporter gene. Most importantly, we demonstrate that both the psoralen cross-link and the purine-motif triplex remained intact for at least 72 h post-transfection, indicating that such species can persist for an extended period of time in vivo. These findings support the feasibility of an antigene approach for the therapeutic regulation of specific gene expression.
A potentially powerful approach to the treatment of chronic diseases such as cancer involves the development of therapeutics that modulate specific gene expression. Such drugs would ideally target only a single gene and specifically affect the process of transcription, a primary means by which gene expression is controlled in higher eukaryotes. Triplex-forming oligonucleotides (TFOs), typically pyrimidine-rich or purine-rich oligodeoxyribonucleotides that interact through Hoogsteen hydrogen bonding to runs of purine acceptors in Watson-Crick duplex DNA, offer one class of possible gene-regulatory molecules (1 ,2 ). By targeting TFOs to the control regions of specific genes, several laboratories have demonstrated the feasibility of such an `antigene' approach, both in vitro and in vivo (reviewed in 3 ). However, the limited repertoire of DNA sequences recognized by currently defined triplex motifs, typically a homopurine-homopyrimidine sequence at least 10 bp in length, greatly restricts the general utility of this approach.
Transcription in eukaryotes is a multistep process, involving nucleosome displacement, promoter recognition, transcription initiation and elongation by RNA polymerases. Whereas the first three typically involve only small regions of the DNA, often comprising sequences totaling <100 bp, transcription elongation occurs over the entire length of the gene to generate a functional RNA product. For many higher eukaryotic genes, this entails a span of several kilobases. Thus, directing TFOs to inhibit transcription elongation instead of initiation greatly increases the likelihood of identifying a suitable homopurine-homopyrimidine target sequence within a particular gene.
Although transcription elongation inhibition has been demonstrated with unmodified oligonucleotides in vitro, convincing results in vivo most often require formation of a covalent adduct between the TFO and the target DNA duplex (1 ,4 -7 ). This is usually achieved through the use of modified oligonucleotides containing DNA reactive moieties. Among these, psoralen-modified TFOs have been the most investigated, given their commercial availability and their potential to form photoadducts with both strands of the target duplex DNA, yielding interstrand cross-links. Such adducts are extremely effective blockades to progression by RNA polymerases (8 ,9 ) and, at present, offer the best potential for developing a general gene-specific transcription- modulating therapeutic.
Some problems have yet to be solved. For one, TFO-directed psoralen-DNA cross-links can be removed in vivo through a process of DNA repair (6 ,10 ,11 ). Since the noncovalently bound triplex itself is typically insufficient to inhibit elongation by RNA polymerase II, repair of cross-links often results in a loss of transcription inhibition. Here we report the use of a 5' psoralen-modified, 19mer G/T-rich TFO that efficiently inhibits the expression of a luciferase reporter plasmid in vivo. Unlike a previous study (6 ), we observe transcription inhibition persisting for up to 72 h after transfection, suggesting that triplex-directed cross-links remained unrepaired within the cell. Most interestingly, a restriction endonuclease protection assay and Southern blotting showed that not only did the psoralen cross-link survive but also that the triplex remained intact.
Our test plasmid for investigating triplex-directed psoralen cross-links in vivo, pMM2 (Fig. 1 ), was constructed from the pGL2-control reporter vector (Promega) through the sequential cloning of the 415-bp EcoRI (end-filled)-HindIII fragment from pC2ATD19, containing the 377-bp G-less transcription cassette (12 ), into StuI/HindIII-digested pGL2-control, followed by cloning of a 30-bp oligonucleotide containing a 19-bp triplex-forming cassette into the resulting HindIII site. A schematic of pMM2 detailing the SV40 promoter/luc gene transcription unit and the triplex-forming cassette is shown in Figure 1 . Plasmids used in this study were purified by two centrifugations through CsCl-ethidium bromide gradients.
To effect triplex formation, we incubated 0.6 [mu]M pMM2 plasmid and 30 [mu]M PODN 1 for 90 min at 30oC in a 25-[mu]l reaction mixture containing 40 mM Tris-HCl (pH 8.0) and 100 mM MgCl2. A high Mg2+ concentration was used to promote maximal purine-motif triplex formation (14 ). Afterwards, KCl was added to a final concentration of 100 mM, which prevented further purine-motif triplex formation (15 ). For control reactions in which triplex formation was omitted (indicated by
The extent of triplex formation was determined by a restriction endonuclease protection assay (REPA) (15 ). In brief, 1 [mu]g plasmid DNA was digested for 30 min with 10 U EcoNI in a 20-[mu]l reaction volume and then electrophoresed through a 1% agarose gel containing 1* TAE (40 mM Tris-acetate, pH 7.8, 0.2 mM EDTA) and ethidium bromide. Since there are two EcoNI sites within pMM2, one overlapping the triplex-forming cassette and one located 1654 bp downstream, the percentage of plasmid DNA present in the singly-cut species (linearized plasmid, 6491-bp species) indicated the extent of triplex formation.
Psoralen photoadduct formation was initially characterized by denaturing PAGE of labeled DNA fragments containing the triplex-forming cassette (16 ). UV-irradiated pMM2 was digested with either HindIII or BamHI, depending on which strand of the duplex DNA was being investigated (nontranscribed or transcribed strand, respectively) and then end-labeled with the Klenow fragment of DNA polymerase and either [[alpha]-32P]dATP or [[alpha]-35S]dATP. Following digestion with the other restriction endonuclease, the resulting DNA fragments were resolved by electrophoresis through a denaturing 6% acrylamide-0.2% bisacrylamide-50% urea gel in 1* TBE (89 mM Tris-borate, 2 mM EDTA) and visualized by autoradiography. Quantitation of the REPA and denaturing PAGE analyses was by densitometry of photographic negatives and autoradiograms, respectively. Measurements were in arbitrary density units.
Following characterization of the psoralen photoadducts, semiquantitative PCR was used to determine the extent of site-specific cross-linking. PCR amplifications were made of regions within pMM2 that either contained the triplex-forming cassette (`A' region) or did not (`B' region), the latter serving as an amplification control. Locations of these regions are shown in Figure 1 . PCR product A, 800 nt long, was amplified using the oligonucleotide primers A1 d(5'-CCGGGAGGTACCGAGGTCTTACG-3') and A2 d(5'-CCAGGGCGTATCTCTTCATACGG-3'), while PCR product B, 301 nt long, was amplified with primers B1 d(5'-GCTTCTGGGGGCGACCTCTTTC-3') and B2 d(5'-CAATCAAGGCGTTGGTCGCTTCC-3'). PCR analysis of cross-linked plasmids before transfection was performed in a 100 [mu]l reaction volume containing 200 ng pMM2, 10 mM Tris-HCl (pH 9.0 at 25oC), 50 mM KCl, 2.5 mM MgCl2, 0.15% Triton X-100, 125 [mu]M dNTPs, 20 ng of each primer and 5 U Taq DNA polymerase. The samples were subjected to 12 cycles of amplification, including denaturation for 1 min at 94oC, primer hybridization for 1 min at 50oC and DNA polymerization for 1 min at 72oC. PCR products were electrophoresed through a 1% agarose-1* TAE gel containing ethidium bromide and visualized by UV fluorescence. Quantitation was by densitometry. The extent of cross-linking was determined by comparing the ratio of the PCR products (A:B) to that observed in an untreated plasmid alone control reaction (indicated by
HeLa Attardi cells were maintained in MEM-Joklik medium supplemented with nonessential amino acids, 2 mM L-glutamine, streptomycin (100 [mu]g/ml), penicillin (100 U/ml) and 10% calf serum bovine donor in spinner flasks at 37oC. The cell density was adjusted to 4 * 105 cells/ml 24 h before transfection. Transfections were performed by electroporating, at 350 V and 950 [mu]F, 5 * 106 cells in a 400-[mu]l volume containing the above growth medium without serum but with 25 mM HEPES (pH 7.4), 20 [mu]g pMM2 and 10 [mu]g RSV-[beta]-galactosidase vector. After electroporation, the cells were diluted with 5 ml growth medium (containing antibiotics and serum) and incubated at 37oC in the presence of 5% CO2 to allow maximal recovery. Standard bacterial Petri dishes were used to minimize cell adhesion. Aliquots of 2.5 ml were harvested 24, 48 and 72 h after transfection and replaced with an equal volume of prewarmed medium.
Luciferase assays were performed following the manufacturer's instructions (Promega) using a Monolight 2010 Luminometer (Analytical Luminescence Laboratory). All assays were performed within the linear range of the instrument (0-400 000 relative luminometer units). [beta]-Galactosidase activity at different times was measured with a resorufin-[beta]-galactopyranoside substrate (Boehringer Mannheim), incubated at 30oC.
To isolate plasmid DNA from the transfected HeLa cells, ~1 * 106 cells from each time point were first resuspended in 500 [mu]l PBS containing 15 mM MgCl2 and 2 mM CaCl2 and then treated with 10 U DNase I for 30 min at 37oC. This DNase digestion was performed to remove any contaminating plasmid DNA adhering to the cell surface, which we have found significantly interferes with subsequent PCR and REPA/Southern analyses (M. Musso, unpublished observations). Treated cells were pelleted and stored at -80oC until needed. Plasmid and genomic DNA were extracted by resuspending a cell pellet in 100 [mu]l of a buffer containing 10 mM Tris-HCl pH 8.0, 10 mM EDTA, 100 mM NaCl, 0.5% SDS and 100 mg/ml Proteinase K, heating at 55oC for 15 min to denature most proteins, and incubating for 16 h at 37oC with continuous agitation to facilitate protein digestion. After two extractions with a 1:1 mixture of phenol-chloroform and one extraction with chloroform alone, the DNA was ethanol precipitated and resuspended in 50 [mu]l TE.
The extent of psoralen cross-linking remaining after transfection was determined by a PCR assay. The reaction conditions were essentially as described above, with the exception that 5 [mu]l of the extracted DNA mixture was used and that the amplification cycles were increased to 34 for the PCR A product and 19 for the B product. The increased number of cycles was required because of the relatively low amount of plasmid DNA present in these extracts (estimated 1 ng); the difference for the A and B PCR reactions reflected the different amplification efficiencies for these two sets of primers. In all cases, amplifications were well within the range where the amount of PCR product was proportional to the amount of input template (i.e. the `exponential' range).
The extent of triplex DNA remaining after transfection was determined by REPA, with specific visualization of the plasmid DNA by Southern blotting. DNA from 20 [mu]l of the extracted mixture was cleaved for 2 h with 40 U EcoNI, size fractionated on a 1% agarose-1* TAE gel and capillary transferred onto a Nytran Plus (Schleicher & Schuell) membrane. Hybridization was carried out for 16 h at 68oC in a solution containing 6* SSC (90 mM sodium citrate, pH 7.0, 900 mM NaCl), 5* Denhardt's reagent, 0.5% SDS, 100 [mu]g/ml denatured salmon sperm DNA and 2.5 * 106 c.p.m./ml 32P-labeled probe made by random priming of the entire pMM2 plasmid. Afterwards, the hybridized blot was washed at 25oC in 2* SSC and 0.1% SDS until background counts were reduced. Plasmid species were visualized by autoradiography and quantitated by densitometry.
Triplex-directed psoralen cross-links were previously shown to effectively inhibit transcription in vivo (6 ). However, their effectiveness was found to diminish over time, presumably the result of cellular DNA repair. To investigate the parameters that mediate this loss of transcription inhibition, we designed a synthetic template and experimental methods that also allowed quantitative analysis of triplex formation and psoralen cross-linking, both before and after transfection. A schematic of the synthetic template is shown in Figure 1 , while a flowchart of the experimental approach is shown in Figure 2 .
Our experimental protocol involved formation of purine-motif triplexes in vitro, followed by psoralen photoadduct formation after UV irradiation at 365 nm (Fig. 2 ). This approach was chosen because triplex formation in vivo, either through the addition of TFOs to the culture medium or by co-transfecting TFOs with the reporter plasmid, is not very efficient (1 ,2 ). Completion of these initial steps was monitored by quantitating triplex formation using REPA with EcoNI and psoralen cross-link formation by PCR. Examples of these assays are shown in Figure 3 . In the REPA assay, triplex formation was indicated by the appearance of a singly EcoNI cleaved pMM2 species, with a length of 6491 bp. Under our standard reaction conditions (30 [mu]M TFO, 100 mM Mg2+), nearly complete (94%) triplex formation was routinely achieved (Fig. 3 A, lane T). As expected, substitution of 100 mM K+ essentially abolished (<5%) triplex formation (lane K). In both instances, subsequent addition of the missing cation and incubation at 30oC for an additional 90 min resulted in no observable effect on triplex formation (data not shown). Taken together, these data demonstrate efficient triplex formation with pMM2 and the TFO PODN 1.
Given the efficient formation of triplex-directed psoralen cross-links, we examined next whether these adducts would effectively inhibit expression of a specific, targeted gene. In brief, the pMM2 reporter plasmid, containing a psoralen cross-link as determined by either PCR or denaturing PAGE analyses, was transiently transfected together with a pRSV-[beta]-galactosidase reporter vector into HeLa cells by electroporation. After either 24, 48 or 72 h in culture, the transfected cells were collected and whole-cell extracts made. [beta]-Galactosidase and luciferase assays were then performed. Luciferase activity provided a measure of transcription efficiency for the luc gene, which contained a psoralen cross-link within its 5' untranslated region. [beta]-Galactosidase activity provided an independent measure of transfection efficiency. Figure 4 shows the data from a representative experiment. Cells containing either cross-linked (solid bars) or not cross-linked (open bars) plasmids both demonstrated increasing levels of transcription activity for the time periods 24 and 48 h after transfection, with a decreased level of transcription observed at the later time point (72 h). This is understandable, given that some plasmid dilution would be experienced following two rounds of cell division. Comparing the levels of transcription between the two types of transfected cells, we found 16.3% of the normal transcription activity after 24 h, 21.0% after 48 h and 22.4% after 72 h, with cells containing cross-linked templates. These values compare favorably with the levels of psoralen cross-linking determined prior to transfection. Most noteworthy, this inhibition persisted relatively unchanged for the entire 72 h period, with inhibition decreasing only slightly with increasing time of culture. This is quite different from previous studies, which reported substantial loss of transcription inhibition only 48 h after transfection (6 ).
The persistence of psoralen-dependent transcription inhibition could indicate that the responsible psoralen cross-link was maintained through this time period, even though the cell line used was DNA-repair competent. Alternatively, the cross-linked plasmid could have been lost and/or damaged, perhaps during the repair process, thereby losing its ability to serve as a transcription template. To discriminate between these two possibilities, we investigated whether plasmids isolated from the transfected cells retained the psoralen cross-links. Three types of reactions were performed: transfection with TFO and pMM2 but without triplex formation (K), transfection with triplex-directed psoralen cross-links (T) and a pseudotransfection in which pMM2 was introduced into the medium immediately after electroporation (C). These last reactions were included to determine whether any of the PCR products observed could be the result of contaminating extracellular plasmid DNA. In each case, aliquots of the cells were removed at different times and exhaustively treated with DNase I to remove any extracellular plasmid contamination. Plasmid and cellular DNA were purified by protease digestion and multiple phenol extractions. These DNAs were used as templates for PCR amplification, with primer set A indicating the extent of cross-linking and primer set B serving as a control, as indicated above (Fig. 1 ). The number of amplification cycles used depended on the efficiency of amplification by a particular primer set but was kept well within the exponential range for pMM2 in the nanogram range.
As shown in Figure 6 A, at 24 h after transfection, the triplex-containing templates exhibited a reduced A:B ratio (T, 4:13) when compared with the plasmid-alone (P, 13:18) or TFO/plasmid (K, 17:17) controls. This difference in the ratio of PCR products was comparable with that observed prior to transfection (Fig. 3 B). Similar ratios were observed after 48 and 72 h, indicating that the original psoralen cross-links survived intact. Note that no PCR products were observed with either primer set in the extracellular plasmid controls (C). This suggests that the PCR products were solely amplified from intracellular plasmids. Also note that for any reaction set, the absolute quantities of PCR products remained relatively unchanged for the different timepoints. This indicates that the plasmid DNA was not lost from the cells during the time period investigated.
While our data suggests that a triplex-directed psoralen cross-link can effectively inhibit a targeted gene's expression for several days in vivo, these experiments do not formally exclude the possibility that the purine-motif triplex itself was responsible for the observed transcription inhibition. This possibility is supported by two recent findings, one concerning the induction of specific mutations in transfected plasmids by unmodified purine-motif TFOs (29 ) and the other a demonstration of transcription inhibition in vivo by an unmodified G/A-rich TFO (30 ). To test the effectiveness and persistence of transcription inhibition by a noncovalently attached purine-motif TFO, we performed a series of experiments using the TFO ODN 1-N. ODN 1-N has the same nucleotide sequence as PODN 1, though instead of a 5' psoralen moiety it contains a 3' alkylamine modification that has been shown to provide improved intracellular stability (31 ).
As determined by REPA (Fig. 7 A), both ODN 1-N and PODN 1 exhibited comparable triplex-forming capabilities, demonstrating >95% triplex formation in the presence of Mg2+ and <5% triplex formation when 100 mM K+ was also present. As expected, following UV irradiation only templates containing PODN 1 triplexes exhibited any detectable photoadduct formation (Fig. 7 B). Under these experimental conditions >83% of the templates had psoralen cross-links while 8% each had monoadducts or were unmodified. Data from luciferase assays on extracts from HeLa cells transfected with these templates are shown in Figure 7 C. For each template, transcription activity generally increased from 24 to 48 h then decreased from 48 to 72 h. This decrease is consistent with a dilution of template following multiple cell divisions. Templates transfected with oligonucleotides, either complexed as triplexes (T) or separate (K), consistently demonstrated higher levels of transcription when compared with a template alone control (P). This increase may be the result of an improved transfection efficiency in the presence of an increased nucleic acids concentration. Alternatively, the oligonucleotides may scavenge cellular proteins that would otherwise inhibit template function. Comparing triplex-containing (T) and uncomplexed (P) templates, we found that templates containing psoralen cross-links (PODN 1) exhibited 6.1, 11 and 15% of the transcription activity of the uncomplexed templates at 24, 48 and 72 h, respectively. These results compare favorably with those presented earlier (Fig. 4 ) and are indicative of an efficient, triplex-dependent transcription inhibition with limited DNA repair throughout a 72 h period. Most interestingly, transcription inhibition was also observed with the noncovalently bound TFO, ODN 1-N. In this case, triplex-containing templates exhibited 60, 78 and 90% of the transcription activity of the uncomplexed templates at 24, 48 and 72 h. Thus our data would suggest that purine-motif triplexes themselves are capable of inhibiting transcription in vivo, albeit to a lesser extent and for a shorter duration than psoralen cross-linked triplexes.
Using the 5' trimethylpsoralen-modified G/T-rich 19mer oligonucleotide PODN 1, we demonstrated efficient purine-motif triplex formation and interstrand psoralen cross-linking to a unique site positioned within the 5' untranslated region of a luciferase reporter gene. Templates containing this covalent modification exhibited significant inhibition of luciferase expression when transfected into HeLa cells. The transcription inhibition persisted relatively unchanged for up to 72 h after transfection. Analysis of the plasmid DNA recovered from the transfected cells indicated that both the psoralen cross-link and the purine-motif triplex survived intact. These findings illustrate, therefore, that the introduction of a positioned photoadduct can effectively inhibit specific gene expression in vivo for extended periods of time.
Two other laboratories have investigated in vivo transcription elongation inhibition by triplex-directed psoralen cross-links, each with different results. Degols et al. (6 ) used a 5' psoralen-modified T/5-methyldeoxycytosine-rich 15mer targeted to a sequence 240 bp downstream of an SV40 promoter to inhibit expression of a modified lacZ gene. Macaulay et al. (7 ) used a 5' trimethylpsoralen/3' alkylamine-modified T/C-rich 20mer targeted to a sequence within the human cytochrome P450 aromatase gene, which was driven by an immediate-early human cytomegalovirus enhancer/promoter. In the former case, HeLa cells transfected with psoralen-modified plasmids demonstrated a >75% reduction in [beta]-galactosidase activity 24 h after transfection but a <20% reduction after 48 h. This loss of inhibition was shown to result from a repair of the psoralen photoadduct through a mechanism not present in xeroderma pigmentosum cells. In the latter case, MCF-7 breast cancer cells demonstrated an 84% reduction in aromatase activity 72 h after transfection and similar levels after 14 days in culture.
From these studies and our own, several experimental parameters emerge that could be responsible for the observed differences in the persistence of transcription inhibition by triplex-directed psoralen photoadducts. The use of a shorter oligonucleotide (15 versus 19 or 20mer) may allow more facile repair of the photoadduct, as has been observed for the differences in repair efficiencies of psoralenated 10mers versus 30mers in COS cells (11 ). However, all of these oligonucleotides are shorter than the reported region required for nucleotide excision repair (i.e. 22-24 nt 5' and 4-6 nt 3' of a lesion) (26 ). Also, a 22mer G/AT-rich TFO and a 33mer G-rich TFO have been shown to be readily repaired in COS cells (10 ,27 ), suggesting that oligonucleotide length may not be the sole defining characteristic regarding triplex-directed psoralen cross-link repair. Other differences include the use of 4,5',8-trimethylpsoralen, which has been shown to have improved dark binding affinity and increased photoreactivity as compared with unsubstituted psoralen (28 ), and highly efficient formation of psoralen diadducts, which are more difficult to repair without mutation than corresponding monoadducts (10 ). Finally, both experiments demonstrating long-term transcription inhibition used psoralen cross-links positioned upstream of the triplex with respect to the transcription start site. In this case, the elongating RNA polymerase II (and any associated transcription/repair proteins) would first encounter the cross-link and would not be expected to appreciably affect triplex stability downstream.
One parameter that may play an important role in affecting the rate of repair for triplex-coupled psoralen cross-links is the stability of the triplex in vivo. Our data suggest that both the psoralen cross-link and the adjoining triplex remain intact intracellularly for up to 72 h. Thus, maintenance of the triplex may be essential for preventing efficient repair of the adjoining psoralen cross-link. Loss of triplex stability, either through passive (e.g., increased dissociation constant of shorter TFOs, use of nonconsensus base triplets) or active means (e.g., disruption by an elongating RNA polymerase) could render the psoralen photoadduct more amenable to nucleotide excision repair. Ultimately, controlling those parameters that affect triplex stability in vivo may be important in the design of targeted psoralen adducts with defined intracellular stabilities.
Can purine-motif triplexes themselves inhibit transcription in vivo? Several lines of evidence suggest that noncovalently bound triplexes are at best kinetic blockades to elongation by RNA polymerases and would not be expected to provide efficient and lasting transcription inhibition (1 ,4 -7 ). However recent evidence for triplex function in vivo, both in generating site-specific mutations (29 ) and in inhibiting transcription from the murine c-pim-1 promoter (30 ), suggests that an ordinary triplex might be capable of affecting transcription elongation in vivo. We found that a noncovalently bound triplex inhibited transcription to a lesser extent and for a shorter duration than that of a triplex-directed psoralen cross-link (Fig. 7 C). As yet it is not known whether this inhibition is achieved by a direct blockade of RNA polymerase II translocation or rather results from an indirect triplex-dependent inactivation of the template. Further experiments (e.g., locating triplexes outside of the promoter and transcribed regions) will be necessary to determine this effect. These data are also consistent with observations that a covalently bound triplex was considerably more efficient in inducing specific mutations than a triplex alone (11 ,29 ). Taken together, these data indicate that while a psoralen cross-linked triplex may be an efficient and enduring inhibitor of transcription, a noncovalently bound triplex may provide a temporary transcription inhibitor. Such would be especially important if the introduction of relatively permanent interstrand cross-links should have undesired consequences beyond merely inactivating targeted gene expression (e.g., preventing DNA replication).
We thank Michèle Sawadogo for critical reading of this manuscript and Randy J. Legerski, Caroline A. Peterson and Qingyi Wei for materials and technical assistance. This work was supported by a grant from the Welch Foundation (G-1199).
REFERENCES
Return