The repair of point mutations can be directed by modified single-stranded DNA oligonucleotides and regulated by cellular activities including homologous recombination, mismatch repair and transcription. Now, we report that DNA replication modulates the gene repair process by influencing the frequency with which either DNA strand is corrected. An SV40-virus-based system was used to investigate the role of DNA synthesis on gene repair in COS-1 cells. We confirm that transcription exerts a strand bias on the gene repair process even when correction takes place on actively replicating templates. We were able to distinguish between the influences of transcription and replication on strand bias by changing the orientation of a gene encoding enhanced green fluorescent protein relative to the origin of replication, and confirmed the previously observed bias towards the untranscribed strand. We report that DNA replication can increase the level of untranscribed strand preference only if that strand also serves as the lagging strand in DNA synthesis. Furthermore, the effect of replication on gene repair frequency and strand bias appears to be independent of certain mismatched base pairs and oligonucleotide length.
Targeted nucleotide exchange, the underlying mechanism of gene repair, can direct the alteration of single bases in biological systems including cell-free extracts, Escherichia coli, Saccharomyces cerevisiae, mammalian cells in culture and a range of animal models including mouse, rat and dog (Alexeev et al., 2000; Alexeev and Yoon, 1998; Bandyopadhyay et al., 1999; Bartlett et al., 2000; Brachman and Kmiec, 2002; Brachman and Kmiec, 2003; Campbell et al., 1989; Cole-Strauss et al., 1996; Cole-Strauss et al., 1999; Ellis et al., 2001; Gamper et al., 2000; Kmiec, 1999; Kren et al., 1997; Kren et al., 1999; Liu et al., 2001; Moerschell et al., 1988; Parekh-Olmedo et al., 2001; Rando et al., 2000; Rice et al., 2001; Santana et al., 1998; Vasquez et al., 2001; Yamamoto et al., 1992a; Yamamoto et al., 1992b; Yoon et al., 1996). Now, efforts are under way to elucidate the mechanism of this dynamic process so that correction efficiencies can be stabilized and eventually raised to therapeutically useful levels. Genetic studies in yeast have revealed that DNA repair proteins play regulatory roles in oligonucleotide-directed repair (Brachman and Kmiec, 2003; Liu et al., 2002a). Additionally, transcriptional activity of the target gene has been found to provide a more open chromatin conformation, which enables a higher frequency of gene repair (Igoucheva et al., 2003; Liu et al., 2002b). A strand bias has repeatedly been observed in which the untranscribed strand is corrected more efficiently (Igoucheva et al., 2001; Igoucheva et al., 2003; Liu et al., 2001; Liu et al., 2002b; Nickerson and Colledge, 2003) (Y. Hu, L. Liu, L. Ferrara and E.B.K., unpublished) and lends further support for a productive role of transcription in the gene repair process. This transcriptional strand bias is probably due to the availability of the untranscribed strand for oligonucleotide alignment, in contrast to the transcribed strand, which is more encumbered by the transcriptional machinery. Furthermore, experiments conducted in vitro indicate that RNA polymerase is able to displace the oligonucleotide from the transcribed strand, reducing its potential for directing a gene repair event (Liu et al., 2002b).
Transcriptional activity alone, however, does not always dictate the strand bias of correction. For example, when targeted nucleotide exchange takes place on the S. cerevisiae CYC1 gene, a gene involved in cytochrome c metabolism, a repair bias is observed that favours the transcribed strand (Brachman and Kmiec, 2003; Yamamoto et al., 1992b). Similar results have now been reported in two unrelated mammalian systems (Bertoni et al., 2003; Bertoni and Rando, 2003; Sorensen et al., 2003), suggesting that processes other than transcription can impact strand bias. In the current work, we examine the possibility that DNA replication influences the process of gene repair strand preference.
Support for this notion comes from important studies in prokaryotes in which DNA replication has already been shown to dominate the strand bias phenomenon (Ellis et al., 2001; Zhang et al., 2003). Recombination within the gal operon of E. coli is higher when oligonucleotides complementary to the lagging strand of replication are used. galK targeting revealed that the bias for the lagging strand remained intact even when the gal operon was inverted, reversing the strands of replication while maintaining the strandedness of transcription. In E. coli, this process is RecA independent (Li et al., 2003), differing mechanistically from the eukaryotic systems, in which the RecA analogue Rad51 appears to be a crucial component of the reaction pathway (Liu et al., 2002a; Lundin et al., 2003; Thorpe et al., 2002; Yanez and Porter, 1999). Nevertheless, these results prompted us to evaluate the role of DNA replication in mammalian cells using the well-characterized SV40 replication system.
Within the cell, the 5.2 kb SV40 genome is contained in a circular duplex minichromosome with a nucleoprotein structure analogous to chromatin. Replication initiates at a well-characterized viral origin through the specific interaction of the viral protein, large tumour (T) antigen (Deb and Tegtmeyer, 1987). Replication proceeds bidirectionally, using both leading- and lagging-strand modes of synthesis. The SV40 system was tailored for our experiments by creating four expression vectors (pSK128-eGFP-1, pSK128-eGFP-2, pSK128-eGFP-3 and pSK128-eGFP-4) derived from pSK128, a Bluescript plasmid containing an SV40 origin of replication (Simmons et al., 1998). A mutant enhanced green fluorescent protein (eGFP) gene was ligated into the pSK128 vector in both orientations and on different sides of the SV40 origin of replication so that the effect of either replication polarity and/or transcriptional status on gene repair could be monitored.
Our plan was to use the SV40 system in order to evaluate the effect of DNA replication on the strand bias of gene repair. By varying the orientation of the eGFP gene within the SV40 viral vector, correlations can be made between strand bias as a function of leading and lagging strand (synthesis) and/or transcribed and untranscribed strand (expression). We find that DNA replication accentuates the frequency of gene repair by directing a strand bias of correction that favours the lagging strand.
Materials and Methods
Cell line and culture conditions
The COS-1 cell line was obtained from ATCC (American Type Culture Collection, Manassas, VA) and was grown in Dulbecco's modified Eagle's medium (DMEM) with 4 mM L-glutamine, 4.5 mg ml–1 glucose, 3.7 mg ml–1 sodium bicarbonate, supplemented with 10% foetal bovine serum at 37°C in 10% CO2.
Plasmid DNA constructs
A single point mutation was created in the eGFP gene of the pEGFP-N3 plasmid (Clontech Laboratories) at codon 67 (TAC→TAG) using a QuikChange Mutagenesis kit (Stratagene). The eGFP cassette including a cytomegalovirus (CMV) promoter was inserted into plasmid pSK128 (a gift from D. Simmons, University of Delaware, Newark, DE) in four different orientations surrounding the origin of replication. The eGFP gene and CMV promoter were PCR amplified from the pEGFP-N3 vector (both wild-type and mutant eGFP) using primers designed to add restriction sites suitable for subcloning the fragment into the pSK128 vector [Invitrogen high-fidelity polymerase chain reaction (PCR) supermix]. SpeI and BstXI restriction sites were added to the eGFP-1 and eGFP-3 constructs and Eco10 and ClaI restriction sites were added to eGFP-2 and eGFP-4 with the following primer sets:
Following PCR amplification, the fragments were isolated, digested with the appropriate restriction enzymes (New England Biolabs) and gel purified. The inserts were then ligated into the pSK128 vector. Plasmids were sequenced to confirm nucleotide alterations.
eGFP3S/72T and eGFP3S/72NT are single-stranded DNA oligonucleotides 72 nucleotides long containing three phosphorothioate linkages at each terminus. eGFP3S/72T targets the transcribed strand of the eGFP gene, and eGFP3S/72NT targets the untranscribed strand. These oligonucleotides are entirely complementary to the eGFP gene except for one centrally located nucleotide designed to convert the TAG stop codon in the eGFP chromophore region to TAC, encoding tyrosine and resulting in a gene that expresses functional protein. eGFP-TAT3S/72T and eGFP-TAT3S/72NT are identical to eGFP3S/72T and eGFP3S/72NT apart from the centrally located base, which, in these cases, was designed to convert the stop codon to TAT (tyrosine). Oligonucleotides eGFP3S/44T, eGFP3S/44NT, eGFP3S/24T and eGFP3S/24NT are analogous to eGFP3S/72T and eGFP3S/72NT except that they varied in length, being 44 (base pairs 14-57 of the 72-mer oligonucleotide) and 24 (base pairs 24-47 of the 72-mer oligonucleotide) nucleotides long, respectively. Hyg3S/74T, a noncomplementary oligonucleotide was used as a nonspecific targeting control (Liu et al., 2001).
Correction of the eGFP gene
Cells were grown to 80% confluence, trypsinized and resuspended in serum-free medium. Square-wave electroporation (BTX ECM 830; 4 mm gap cuvette, two pulses, 200 V, 20 milliseconds) was used to introduce 5 μg plasmid and 10 μg oligonucleotide into the COS-1 cells (2×106 cells in a 100 μl volume of serum-free medium). The electroporated cells were resuspended in complete medium, seeded in a six-well plate and allowed to recover for 40 hours after oligonucleotide treatment. The cells were then trypsinized and resuspended in buffer [PBS, 0.5% bovine serum albumin (BSA), 2 mM EDTA pH 8.0, 2 μg ml–1 propidium iodide (PI)] and assayed using a FACScaliber flow cytometer (Becton Dickinson, Rutherford, NJ). Correction was calculated using the CellQuest and GFP/PI programs. More specifically, the program was set for the appropriate cell size (forward scatter versus side scatter) and the population of single cells was gated for analysis. Using the negative control (minus PI, minus GFP), the background fluorescence was set by positioning the cells in the tenth decade of the dot plot by adjusting the voltage for the FL1 (GFP) and FL2 (PI) channels. The machine composition was then set for multi-fluorochrome experiments using a GFP control sample containing no PI and increasing the compensation to bring the signal toward the FL1 parameter. Finally, the last control (PI and no GFP) was used to increase the compensation to bring the signal toward the FL2 parameter. Samples of 50,000 cells each were analysed and cells that were GFP positive and PI negative were scored as corrected cells. Transformation efficiency was determined by electroporating wild-type eGFP plasmids identical in gene orientation relative to the SV40 origin into COS-1 cells and measuring the level of eGFP expression. The correction efficiency of mutant eGFP plasmids was then calculated by dividing the percentage of fluorescent cells resulting from oligonucleotide treatment by the percentage of fluorescent cells electroporated with the wild-type gene. This control was part of every experimental design and run in every experiment.
DNA replication affects, but does not change, the preferential correction of the untranscribed strand
In order to simplify the terminology in this paper, we use the terms `targeted nucleotide exchange' (TNE) and `gene repair' interchangeably. The basic experimental system contained a mutant eGFP cassette under the control of a CMV promoter inserted into a plasmid, pSK128, adjacent to the SV40 origin of replication (Fig. 1). The mutation in the eGFP gene is a single base substitution (TAC→TAG), creating a stop codon. Square-wave electroporation was used to introduce the mutant eGFP plasmid and oligonucleotide into COS-1 cells. The single-stranded DNA oligonucleotides are composed of a single, centrally located noncomplementary base and three phosphorothioate linkages at each terminus to provide resistance to nuclease degradation. Correction, as directed by the oligonucleotide, enables a phenotypic readout of conversion through expression of the enhanced GFP gene. Correction events were measured after a 40 hour recovery period using fluorescence-activated cell sorting (FACS) analysis.
A series of eGFP reporter constructs was created in order to examine the influence of replication on the previously established strand bias of correction, presumably determined by transcriptional activity. Plasmid pSK128 served as a host for the new constructs that differ in the orientation of the mutant eGFP gene relative to the SV40 origin of replication (Fig. 2). Essentially, the leading and lagging strands are reversed relative to the transcribed or untranscribed strand in this panel of constructs. For example, eGFP-1 and eGFP-2 differ in the combination of strands that function as the leading or lagging and transcribed or untranscribed; in eGFP-1, the leading strand is paired with the untranscribed strand etc. This panel enables us to assess the impact of transcription and replication on gene repair either independently or jointly. COS-1 cells, which express T antigen, are able to support replication of a SV40-based plasmid and thus were used as the host cells in these experiments.
The replication assay was initialized by the transformation of the appropriate plasmid and the oligonucleotide directed to either the transcribed or the untranscribed strand of the mutant eGFP gene. The correction reaction was carried out for 40 hours and the converted cells identified by FACS. As shown in Fig. 3A, the results indicate that each eGFP plasmid harbours a correction bias favouring the untranscribed DNA strand, further supporting previous observations that the untranscribed strand is the preferred target in the gene repair reaction. The level of transcriptional bias varied between 1.5 and 3 times, and no replication bias is obvious on initial inspection; plasmids eGFP-1 and eGFP-4 are more readily corrected on the leading strand, whereas the lagging strand is corrected more frequently in plasmids eGFP-2 and eGFP-3. The data indicating a bias towards the untranscribed strand on each plasmid target was then re-analysed and recalculated by dividing the correction obtained with the NT oligonucleotide by the correction obtained with the T oligonucleotide (Fig. 3B). Plasmids eGFP-1 and eGFP-4 are similar in construction, in that the untranscribed strand correlates with the leading strand and plasmids eGFP-2 and eGFP-3 are similar in that the untranscribed strand correlates with the lagging strand of replication (Fig. 2). When Student's t test was conducted comparing the untranscribed strand bias found for plasmids eGFP-1 and eGFP-4 with the preference found on plasmids eGFP-2 and eGFP-3, the correction values and their standard deviations reveal a statistically significant difference, with P values of 0.002 (Fig. 3B). In short, when the untranscribed strand is also the lagging strand, as in plasmids eGFP-2 and eGFP-3, a higher level of correction is observed than when the untranscribed strand is also the leading strand, as in plasmids eGFP-1 and eGFP-4. Because the major difference between these pairs of plasmid targets is that the replication strand is reversed (the identity as a untranscribed strand is maintained), we believe that the DNA replication process itself impacts the strand bias. The highest frequency of repair occurs when the lagging strand of replication is also the untranscribed strand of the gene.
Changing the mismatch created by the oligonucleotide does not alter the favourable NT/lagging strand bias
Both oligonucleotides (eGFP3S/72NT and eGFP3S/72T) are designed to convert the TAG stop codon to TAC, encoding tyrosine, but eGFP3S/72NT creates a G-G mismatch with the untranscribed strand, whereas eGFP3S/72T creates a C-C mismatch with the transcribed strand. To evaluate the potential impact of such a discrepancy on the factors that influence the repair bias outlined above, two new oligonucleotides were designed that direct the conversion of the mutated TAG codon to TAT. These oligonucleotides, eGFP-TAT3S/72NT and eGFP-TAT3S/72T, create a G-A and a C-T mismatch, respectively, with the designated strand (NT or T). Plasmids, pSK128-eGFP-1 and pSK128-eGFP-3, were targeted with these T and NT oligonucleotides to determine whether the replication influence on transcription bias remains, regardless of a change in base mismatch. The same strategy was used to measure the effects of various mismatches on a similar process in E. coli (Li et al., 2003). As shown in Table 1, the relationship between transcription and replication strands, favouring the NT/lagging strand combination in gene repair frequency, persisted in spite of the alternative mismatch created by eGFP-TAT3S/72NT or eGFP-TAT3S/72T.
Oligonucleotide length does not affect the NT/lagging strand bias but does affect overall frequency
One of the most reproducible effects in gene repair is the observation that the length of the oligonucleotide influences the frequency of the reaction (Brachman and Kmiec, 2002; Ellis et al., 2001; Igoucheva et al., 2001; Liu et al., 2001; Zhang et al., 2003). In general, the longer the oligonucleotide, the higher the correction frequency obtained, independent of the preference for the untranscribed strand. Thus, strand bias and length dependence were tested in this system using the standard 72-mer oligonucleotide and two shorter derivatives – a 44-mer and a 24-mer oligonucleotide. As shown in Fig. 4, a length dependence was found, confirming other reports, with the standard 72-mer yielding a higher level of correction events than the 44-mer; the 44-mer was more effective than the 24-mer. Importantly, however, the relationship between T/lagging, NT/leading, T/leading and NT/lagging strands using plasmids pSK128-eGFP-1 and pSK128-eGFP-3 remained the same for all oligonucleotides.
The gene repair reaction is initiated with the binding of an oligonucleotide specific to the target sequence and the recruitment of cellular proteins, which then catalyse nucleotide exchange. We believe that the rate-limiting step of this reaction is the hybridization of the oligonucleotide to the target DNA, a step largely regulated by the availability of the DNA target. For example, an actively transcribed region is more accessible to oligonucleotide binding because it provides an environment in which the chromatin structure surrounding the gene is loosened (Igoucheva et al., 2003; Liu et al., 2002b). A decondensed chromatin structure is known to be more prone to homologous DNA pairing events necessary for both recombination and targeted nucleotide exchange (Kotani and Kmiec, 1994). The importance of transcription in gene repair was first defined in S. cerevisiae using an inducible promoter on the target gene. Under transcriptional repression, a low level of correction was seen, whereas a higher level of gene correction was measured when the template was active (Liu et al., 2002b). Experiments in mammalian cells with a LacZ gene under the control of the Tet-on and Tet-off promoter confirmed this observation as an actively transcribed gene was again found to be repaired at a higher frequency than inactive targets (Igoucheva et al., 2003). DNA replication can also lead to an altered or perhaps even disrupted chromatin structure, which could create an environment that is more amenable for oligonucleotide pairing and gene repair (Morales et al., 2001).
The influence of transcription on strand bias has been well documented; oligonucleotides designed to hybridize to the untranscribed strand of the target result in a higher level of correction than when the transcribed strand is similarly targeted (Y. Hu, L. Liu, L. Ferrara and E.B.K., unpublished) (Igoucheva et al., 2001; Igoucheva et al., 2003; Liu et al., 2001; Liu et al., 2002b; Nickerson and Colledge, 2003). The basis for this transcriptional bias has not been fully elucidated but one explanation was suggested by in vitro modelling experiments demonstrating that an active T7 phage RNA polymerase can displace the oligonucleotide from the transcribed strand of DNA but not the untranscribed strand (Liu et al., 2002b).
Strand bias favouring the untranscribed strand is not universal, however, because studies reported by Sherman and colleagues (Yamamoto et al., 1992b) with the CYC1 gene of S. cerevisiae revealed a preference for the template or transcribed strand of DNA. Recently, we re-examined gene repair on this gene (Brachman and Kmiec, 2003), targeting five different cyc1– strain variants containing mutations at a codon downstream of the initial mutation site. Four out of the five cyc1– strains revealed a preferential targeting mode for the untranscribed strand but one strain, YMH53, paralleled the early studies, revealing a transcribed strand bias. Further examples of a template preference have appeared in episomal targeting of the LacZ gene in CHO cells in which two mutations approximately 900 bp apart exhibited opposite strand biases (Sorensen et al., 2003). In addition, chromosomal targeting of the mouse dystrophin gene yielded the more commonly observed untranscribed strand bias in vitro, but in vivo studies revealed a transcribed strand bias (Bertoni et al., 2003; Bertoni and Rando, 2003). Studies such as these and others (Lu et al., 2003) make it clear that, although DNA transcription might be a major determinant of strand bias, it is not the only process controlling this choice in the reaction pathway.
Up to this point, the influence of DNA replication on targeted nucleotide exchange using modified oligonucleotides had not yet been examined in detail. In a related study, however, DNA replication in prokaryotes was found to have a significant effect when a 70-mer unmodified oligonucleotide was used to direct homologous recombination in E. coli (Ellis et al., 2001). Examination of oligonucleotide-mediated recombination at five genes on the chromosome indicated a preference correlating not to transcription but to the lagging strand of DNA replication. More recent data from the same group (Li et al., 2003) indicate an absolute relationship between DNA replication and strand bias, and these workers conclude that transcription plays no part in determining strand bias in an E. coli system. Such a restriction might be unique to E. coli, because yeast, and now mammalian cell, strand bias in gene repair appears to be governed by multiple processes (Bertoni and Rando, 2003; Brachman and Kmiec, 2003; Igoucheva et al., 2001; Igoucheva et al., 2003; Liu et al., 2001; Liu et al., 2002b; Lu et al., 2003; Nickerson and Colledge, 2003; Sorensen et al., 2003). This might reflect an increased complexity in eukaryotes compared with prokaryotes or the fact that the E. coli reaction takes place in the absence of the DNA recombinase RecA. By contrast, eukaryotes display a dependence on the RecA homologue Rad51 (Liu et al., 2002a; Lundin et al., 2003; Thorpe et al., 2002; Yanez and Porter, 1999).
Here, we tested a bank of target plasmids containing various combinations of transcribed and untranscribed strands of the mutant eGFP gene in a construct in which they were either the leading or the lagging strand of replication. This system has allowed us to contrast the impact of replication and the influence of transcription. Independent of the replication polarity of each strand, the target plasmids were preferentially corrected on the untranscribed strand (Fig. 3A). However, the degree of strand bias appears to be influenced by both transcription and replication, processes that can occur simultaneously within the same region of the chromosome. A statistically significant increase in the untranscribed strand bias occurs when it also serves as the lagging strand of the replication process (Fig. 3B). The impact of a replication bias presents itself most obviously as an enhancer of the transcriptional bias inherent in the correction process of the mutant eGFP gene. One effect of the lagging strand serving as the template strand for gene repair is that it can also reduce the numerical differences between the strand bias ascribed to the transcribed and untranscribed strands. The transcriptional bias seen in plasmids eGFP-1 and eGFP-4 is relatively small compared with that seen on other target genes and cell lines. The reason for this reduced transcriptional bias is probably a competitive effect between transcription and replication, because the link between the untranscribed and lagging strands is uncoupled and each might effectively `compete' for the oligonucleotide. It should be realized that our experiments are carried on a template with a reporter gene that is heavily transcribed from a strong promoter. `Natural' genes might not be transcribed at such a high rate and thus replication might exert an even greater influence on strand bias under these conditions.
The influence of mismatch specificity in gene repair has also been examined most recently in a yeast system (Brachman and Kmiec, 2003) but the formulation of a complete mismatch hierarchy is problematic because assay readout is dependent on transcription of the gene, limiting mismatch choices (Li et al., 2003). So, mismatches cannot be tested at random or in such a way that genetic readout or gene function is disrupted. In our system, the stop codon within the chromophore region of the mutant eGFP gene must be converted to a tyrosine in order to maintain the production of a functional protein. The NT oligo which directs conversion to a TAT produced a slightly higher level of correction than the NT oligo directing conversion to a TAC codon. This might indicate that a G/A mismatch is favoured over a G/G mismatch, but these do not exhibit a statistical difference. Although only two mismatch types were examined on each DNA strand, it appears in this restricted system that mismatch discrimination plays no significant role in determining strand bias compared with the forces exerted by either DNA transcription or replication or both (Table 1).
Finally, oligonucleotide length was examined by comparing correction frequencies obtained using a 72-mer, a 44-mer, or a 24-mer oligonucleotide directed to either the transcribed or untranscribed strand. An oligonucleotide length response was observed on the overall frequency of gene repair in that the 72-mer directed more correction events than the 44-mer, whereas the 24-mer led to the fewest conversion events. Consistent with our previous observation (Liu et al., 2002b), each of these oligonucleotides exhibited a transcription bias favouring the untranscribed strand. Furthermore, the enhanced bias effect of the lagging strand of replication when identical to the untranscribed strand was also observed for each oligonucleotide length variant (Fig. 4).
DNA replication can now be added to a list of cellular processes that influence the process of gene repair. Transcription was found previously to have a decisive effect on strand bias and could even mask the influence of DNA replication on strand preference. However, we do find a role for DNA replication in creating a strand bias of gene repair, increasing the frequency of repair when the lagging strand of the replication fork is the untranscribed strand of the gene. Other influences on the strand bias of correction will probably appear as mechanistic studies continue.
We thank D. Simmons for providing the pSK128 plasmid used in this work and also for helpful discussions. We also thank fellow laboratory members for useful critiques of the data and the conclusions drawn. And we are most grateful to S. Engstrom for manuscript preparation and editorial assistance. Support from NIH (R01 CA89325) is acknowledged.
- Accepted April 6, 2004.
- © The Company of Biologists Limited 2004