High Fidelity of Internal Strand Transfer Catalyzed by Human Immunodeficiency Virus Reverse Transcriptase*

A system to study the fidelity of internal strand transfer events was constructed. A donor RNA, on which reverse transcriptase (RT)-directed DNA synthesis was initiated, shared homology with an acceptor RNA, to which DNAs initiated on the donor could transfer. The homology occurred over a 119-base internal region of the donor which coded for the N-terminal portion of the α-lac gene. Polymerase chain reaction (PCR) was used to amplify DNA synthesis products. The PCR products were then digested with PvuII and EcoRI and ligated into a vector which had this same region excised. Transformed Escherichia coli were screened for the ability to produce a functional β-galactosidase protein by blue-white phenotype analysis with white colonies scored as those with errors in α-lac. Products synthesized on the donor were used to assess the error rate of human immunodeficiency virus-RT while products transferring to and subsequently extended on the acceptor (transfer products) were used to monitor transfer fidelity. Human immunodeficiency virus-RT made approximately 1 error per 7500 bases copied in the assay. Nucleocapsid protein (NCp), although stimulating strand transfer 3-fold, had no effect on RT fidelity. Transfer products in the absence of NCp had essentially the same amount of errors as donor-directed products while those produced with NCp showed a slight increase in error frequency. Overall, strand transfer events on this template were highly accurate. Since experiments with other templates have suggested that transfer is error prone, the fidelity of strand transfer may be highly sequence dependent.

can result in the integration of genetic information from both genomic copies. The resulting provirus codes for new genomic RNA that is a chimera of the parental genomes and, therefore, not identical to either.
Strand transfer occurs when DNA synthesized on one template is translocated to another region on the same or a different template. Two such events, the transfer of minus and plus strand strong-stop DNAs, are an integral part of retroviral replication (for a review, see Ref. 12). These DNAs are initially synthesized at the 5Ј ends of the viral RNA (Ϫ strong stop) or minus strand DNA (ϩ strong stop). Subsequently, each is transferred to a homologous region located at the 3Ј end of the respective templates. As a consequence of the diploid nature of retroviruses, strand transfer from internal regions of the RNA genome can also occur during the synthesis of minus strand DNA. Such recombinational events are proposed to occur by a "forced copy choice" mechanism (13,14). The forced copy choice model postulates that some viral genomic RNAs are truncated and, therefore, unable to produce a completed copy of minus strand DNA. When DNA synthesis reaches the end of the truncated RNA, the nascent DNA is "forced" to switch to the homologous region of the second copy of genomic RNA within the diploid virion to complete minus strand synthesis. Others have shown that recombination of this type may not require broken RNAs (15). In such a situation, the template switch may not always be forced. A slightly modified version of this model termed "copy choice" could describe all recombinations occurring during minus strand DNA synthesis. Note also that recombination may occur during the synthesis of plus strand DNA as proposed in the "strand displacement-assimilation model" (16,17). Although reports have suggested that plus strand recombination may occur (18), other results have indicated that recombination during the synthesis of minus strand DNA is the predominant pathway (9,19).
Internal strand transfer has been shown to occur both in vivo (20 -22) and in vitro (23,24). Models for strand transfer have been proposed although specific details of the mechanism remain unclear (10,24,25). It has been suggested that strand transfer may be error prone (26,27). This is based on the finding that HIV-RT tends to incorporate additional non-template-directed bases at the 3Ј end of a growing DNA strand after reaching the 5Ј terminus of the template (26 -28). Such events could be particularly relevant to strong stop transfers or forced copy choice type transfers, all of which can happen at template termini. It has also been shown, using an in vitro system designed to test the fidelity of internal transfer within a hypervariable region of the nef gene, that strand transfer was more error prone than RNA-directed DNA synthesis (29). In contrast, results using an in vivo system based on Moloney murine leukemia virus suggested that strand transfer was highly accurate (30). We investigated the fidelity of strand transfer using a system in which transfer occurred between RNAs containing homologous sequences from the ␣ region of ␤-galactosidase. Reverse transcriptase-derived DNA products were PCR-amplified and tested in an ␣-complementation assay. Errors occurring during RNA-directed DNA synthesis or strand transfer were scored based on the inability of mutated products to complement ␤-galactosidase activity, resulting in white rather than blue colonies in the assay. Results showed that strand transfer and RNA-directed DNA synthesis were of approximately equal fidelity. The presence of viral nucleocapsid protein (NCp) enhanced strand transfer but had little effect on fidelity. This suggests that transfer on this substrate is a highly accurate process. Taken together with results from others, our work implies that the fidelity of strand transfer may be highly sequence dependent.

Materials
Recombinant HIV-RT, having properties described (31), was graciously provided to us by Genetics Institute (Cambridge, MA). This enzyme had a specific activity of approximately 40,000 units/mg (1 unit of RT is defined as the amount required to incorporate 1 nmol of dTTP into nucleic acid product in 10 min at 37°C using oligo(dT)-poly(rA) as primer-template). As we have previously reported, the enzyme preparations contained very low levels of single strand nuclease activity (24). We found that this activity could be inhibited by including 5 mM AMP in the assays. The AMP, at this concentration, did not affect the polymerase or RNase H activity of the RT (data not shown). Aliquots of HIV-RT were stored frozen at Ϫ70°C and a fresh aliquot was used for each experiment. HIV-1 NCp was from Enzyco (Denver, Colorado). This 55-amino acid protein corresponded to the p7 portion of HIV-1 MN (32). RNase (DNase-free), T3 RNA polymerase, calf intestinal phosphatase, rNTPs, and dNTPs were obtained from Boehringer Mannheim Biochemicals. T4 polynucleotide kinase, Sequenase, Klenow polymerase, and T4 DNA ligase were from U. S. Biochemical Corp. Taq polymerase and placental RNase inhibitor were from Promega. Restriction enzymes were from Boehringer Mannheim, New England Biolabs, or Life Technologies. Oligonucleotides were synthesized by Genosys Inc. All other chemicals were from Fisher Scientific or Sigma. Radiolabeled compounds were from NEN Life Science Products.

Methods
Strand Transfer and DNA Synthesis Reactions with HIV-RT-Donor template RNA (2 nM, see Fig. 2B), was hybridized to 5Ј 32 P-labeled primer DNA as described below. This template-primer was preincubated for 5 min at 37°C in the presence or absence (as indicated) of 10 nM RNA acceptor template and presence or absence of NCp (2 M final concentration), in 21 l of buffer (see below). Four l of HIV-RT (10 units (approximately 85 nM final concentration)) in 50 mM Tris-HCl (pH 8.0), 1 mM dithiothreitol, and 80 mM KCl (buffer A) was added to initiate the reactions. Reactions included the following reagents at the indicated final concentrations: 50 mM Tris-HCl (pH 8), 80 mM KCl, 6 mM MgCl 2 , 1 mM dithiothreitol, 0.1 mM EDTA (pH 8), 5 mM AMP, 100 M ZnCl 2 , 100 M dNTPs, and 0.4 units/l RNase inhibitor. Reactions were incubated for 40 min at 37°C and stopped by addition of 25 l of 2 ϫ formamide dye (90% formamide, 10 mM EDTA (pH 8.0), 0.1% xylene cyanol, 0.1% bromphenol blue) containing 0.5 g of RNase (DNase free). Samples were heated to 65°C for 10 min to digest the RNA and then for 2 min at 90°C. Samples were electrophoresed on 8% denaturing polyacrylamide gels as described below. Wet gels were exposed to film and transfer or full-length donor-directed product were excised and eluted (24). Some gels were dried and used for product quantitation by phosphoimagery using a Bio-Rad GS-525 PhosphorImager.
RNA-directed DNA Synthesis with the Klenow Fragment of DNA Polymerase I-These assays were performed essentially as described above with the following changes: 1) 5 units of Klenow were used in place of RT; 2) the final [KCl] was 10 mM; 3) AMP was excluded from the reactions.
RNA-DNA Hybridization-For strand transfer reactions the DNA primer was hybridized to donor RNA by mixing primer:RNA transcript at approximately a 5:1 ratio of 3Ј termini in buffer A. The mixture was heated to 65°C for 5 min then slow cooled to room temperature.
Preparation of RNAs-Run-off transcription was done as described in the Promega Protocols and Applications Guide (1989). For the donor template, plasmid pBS⌬MCS, prepared as described below, was cleaved with BglI and T3 RNA polymerase was used to prepare run-off RNA transcripts approximately 189 nucleotides in length. For the acceptor template, plasmid pBS⌬PvuII 1146 was cleaved with PvuII and T3 RNA polymerase was used to prepare run-off transcripts approximately 179 nucleotides in length. The transcription reactions were extracted with phenol:chloroform:isoamyl alcohol (25:24:1) and precipitated with ethanol. The RNA was gel-purified on denaturing polyacrylamide gels, located by ultraviolet shadowing, and recovered as described previously (24). The amount of recovered RNA was determined spectrophotometrically from optical density.
Polymerase Chain Reaction (PCR) and DNA Sequencing-Strand transfer products from reactions performed with acceptor template, or full-length donor-directed products from reactions performed without acceptor, were excised and eluted from denaturing gels as described above. The eluted DNA was amplified by PCR using the following primers: 5Ј-CCTCTTCGCTATTACGCCAG-3Ј and 5Ј-GCTCGAATTCG-CCCTATAGTGAGTC-3Ј. The first is identical to the primer used to (1 min) were performed. Products were extracted, precipitated, and digested as described in Fig. 2B. Fifty units of EcoRI per PCR reaction was used and after 1 h, 2.5 units of calf intestinal phosphatase was added and incubation was continued for an additional hour. The calf intestinal phosphatase was inactivated according to the manufacturer's instructions and the DNA was recovered by precipitation. Thirty units of PvuII for 1 h was used to digest each reaction. The samples were then electrophoresed on an 8% native polyacrylamide gel as described above. Products were located by UV shadowing and recovered as described above, and quantitated by spectrophotometry. These products were ligated into vector pBS⌬PvuII 1146 which had been previously cleaved with PvuII and dephosphorylated with calf intestinal phosphatase, then cleaved with EcoRI. The linear vector was recovered from a 1% agarose gel with a Qiagen gel extraction column. The vector (50 ng) and insert (0.05 pm) were ligated for 16 h at 16°C in a volume of 10 l. Two l of the ligation was used to transform Escherichia coli XL-1 Blue competent cells (Stratagene) according to the manufacturer's protocol. Cells were plated in the presence of 5-bromo-4-chloro-3-indolyl-␤-Dgalactoside and isopropyl-␤-D-thiogalactopyranoside. Colonies were analyzed and scored as blue or faint blue or white. For DNA sequencing, mini-preps were prepared from white or faint blue colonies and the DNA was sequenced using Sequenase (U. S. Biochemical Corp.) according to the manufacturer's protocol. The sequencing primer was 5Ј-GGAAACAGCTATGACCATGA-3Ј.
Construction of Plasmids pBS⌬MCS and pBS⌬PvuII 1146 -Both plasmids were derived from pBSM13ϩ (Stratagene Inc.). Plasmid pBS⌬MCS (see Fig. 2A) was made by cleaving pBSM13ϩ with EcoRI (position 881) and HindIII (position 932) and filling in both ends with Klenow. After ligation with T4 DNA ligase a transformant containing recircularized plasmid which had lost the portion of the multiple cloning site (MCS) between the EcoRI and HindIII (confirmed by sequencing) was expanded and the plasmid was isolated using a Qiagen Mega-Prep kit. Plasmid pBS⌬PvuII 1146 was produced through multiple steps as described in Fig. 1. This plasmid has a 7-base pair (1149 -1155 of pBSM13ϩ) deletion that disrupts the PvuII site at position 1146, leaving a single site for this enzyme at 764.

System Used to Study Strand Transfer-
The templates used to study strand transfer and the plasmids used to make those templates are shown in Fig. 2. The donor RNA (produced from pBS⌬MCS ( Fig. 2A), on which RT initiates DNA synthesis, was primed with a specific 20-nucleotide DNA oligonucleotide labeled at the 5Ј end with 32 P. Extension to the end of the donor would produce a 152-nucleotide product while homologous transfer of the growing DNA to the acceptor (produced from pBS⌬PvuII 1146 ) and subsequent extension yields a 199-base product (Fig. 2B). The region of homology (transfer zone) between the donor and acceptor encompassed 119 bases corresponding to a region of ␣-lac near the N terminus of this protein. Transfer or full-length donor-directed products were isolated from denaturing polyacrylamide gels (see Fig. 3) and used for PCR as described under "Methods." The products were processed as shown in Fig. 2B (see "Methods" for details). The final 119-base pair product, which was dephosphorylated at the EcoRI cleaved end, was ligated into plasmid pBS⌬PvuII 1146 ( Fig. 2A) which had this same fragment excised and was dephosphorylated at the PvuII-cleaved end. The plasmid and insert were dephosphorylated to decrease the likelihood of ligation products consisting of multiple inserts and/or plasmids. Such products would generally produce white colonies due to the orientation of the ligated products and not the sequence of the insert. After transformation, colony color was analyzed with white or faint blue colonies scored as those carrying plas- mids with mutations in the ␣-lac region. We note that not all mutations will be manifested as white or faint blue colonies. Frameshifts, which disrupt the reading frame of ␣-lac, are generally detected while base changes that result in amino acid changes may be detected depending on their effect on ␤-galactosidase activity. Base changes in the wobble base position that do not change amino acids will not be detected.
There are several "trivial" ways that the ␣-lac regions of the transformed plasmids could have been disrupted. These include ligation product multimers, T3 RNA, or Taq polymerasederived errors, or contaminating nuclease activities that modified the insert or vector during preparation. In order for the assay to be sensitive enough to calculate HIV-RT-derived errors, it is important that the sum of all these errors be considerably less than the errors derived from HIV-RT. Background mutant colonies resulting from restriction enzyme digests and ligation procedures were estimated by cleaving the PvuII-EcoRI insert from plasmid pBS⌬PvuII 1146 using the protocol described in Fig. 2B, and ligating this insert back into the vector as described. Insert prepared in this manner were not subject to errors arising from T3 RNA polymerase, Taq, or RT. Approximately 20,704 colonies were analyzed using this approach and 88 of these were white or faint blue (data not shown). The mutant colony frequency (mutants/total colonies) from this data was 4.3 ϫ 10 Ϫ3 . The error frequency per base was calculated by dividing this value by 99, the number of bases between the primers used for PCR in Fig. 2B. This is appropriate since the other 20 bases of the 119-base insert, although originally derived from RT in the protocol, would be specified in the insert by the PCR primer binding near the EcoRI site. The background error frequency was 4.3 ϫ 10 Ϫ5 or 1 error per about 23,000 bases. Note that this is about 7-fold lower than the frequency after HIV-RT synthesis (Table I, second column). However, products produced from HIV-RT were also subject to T3 and Taq polymerase errors. To estimate the contribution of these enzymes to the error frequency we performed RNA-directed DNA synthesis using the Klenow fragment of E. coli DNA polymerase I. Although this enzyme is a DNA-dependent DNA polymerase, it can also use RNA as a template (34). The fidelity of this enzyme on RNA was reported to be relatively high, with an estimated error frequency in the 10 Ϫ6 range (35). Also, using Klenow in an ␣-complementation assay similar to the one employed in this work, Ji and Loeb (8) found that Klenow was at least 4-fold more accurate than HIV-RT. When Klenow was used to reverse transcribe the donor RNA, an error frequency per base of about 1.5 ϫ 10 Ϫ4 was calculated (see Table I). This frequency is about 50% that calculated with HIV-RT. Therefore, at least half of the errors observed when HIV-RT was used were derived from RT.
Nucleocapsid Protein Stimulates Strand Transfer-An autoradiogram from a typical strand transfer experiment is shown in Fig. 3. Full-length donor-directed (F) or transfer products (T) were excised from gels and processed as described under "Methods." Donor-directed products were taken from assays without acceptor template. The presence of NCp (2 M) in the reactions stimulated strand transfer about 3-fold. The efficiency of transfer, defined as the amount of transfer product divided by the sum of transfer plus full-length products ϫ 100 ((T/(T ϩ F)) ϫ 100) was about 12% (average of three experiments) without and 36% (average of two experiments) with NCp. Note also that when NCp was used, several of the pause sites (sites on the template where premature termination occurs) evident in the absence of acceptor faded when acceptor was added. This suggests that the paused products are "chased" into transfer products by binding to the acceptor and subsequently being extended (24). The "chasing" was not observed without NCp although transfer clearly occurred. However, in the presence of higher amounts of acceptor template (40 nM), several of the pause sites decrease in intensity and the level of transfer products increases even in the absence of NCp (data not shown). Others have also shown that the level of strand transfer is proportional to the amount of acceptor (36). It is likely that chasing was not detected in Fig. 3 due to the low level of transfer products without NCp and not because it only occurs in the presence of NCp. The fact that several different paused products appear to transfer suggests that strand transfer occurs from several locations on the template.

Transfer and Donor-directed Products Showed Approximately the Same Level of Errors in the Presence or Absence of
NCp-Transfer and donor-directed products were processed as described in Fig. 2B. Analyses from three independent experiments are shown in Table I. In each experiment, 1500 to several thousand colonies derived from a given DNA source were scored. Error frequencies for transfer and donor-directed products produced using standard conditions did not vary significantly. As was previously noted, products produced with Klenow had significantly less errors. We also performed assays using suboptimal concentrations of dATP during HIV-RT synthesis. The error frequency increased about 10-fold when 1 M dATP was used as opposed to 100 M in the standard assay (data not shown). These results indicate that a decrease in fidelity can be detected by the assay. The only notable deviation for HIV-derived products was a small increase in the error frequency for transfer products in the presence of NCp. The third column shows the results after subtraction of background (see definition in table legend) or subtraction of the error frequency obtained using Klenow (in parentheses). These values reflect the highest and lowest error frequencies, respectively, for HIV-RT in the assay. The first value assumes an error frequency of zero for RNA polymerase and Taq while the second assumes a zero value for Klenow. Based on these values the detectable error frequency of RT in this assay was between 1.3 ϫ 10 Ϫ4 and 2.4 ϫ 10 Ϫ4 or 1 error per approximately 4200 -7700 bases. This value is close to that obtained for RNAdirected DNA synthesis by Ji and Loeb (8), but is significantly higher than values from others (6,29). Also shown in Table I are calculations for the transfer frequency. These numbers a Full-length refers to DNA products extended to the end of the donor template while transfer products transferred to, and were extended on the acceptor template. Products derived from synthesis with the Klenow fragment of Pol I were full-length.
b The error frequency per base is defined as the mutant colony frequency/99. Ninety-nine being the number of bases in the region being analyzed for mutations. The mutant colony frequency is the number of white or faint blue colonies/total colonies. Results were from three experiments.
c Background is defined as the error frequency obtained from insertion of a "non-mutated" insert which was homologous to the inserts derived from the different DNA sources by PCR (see "Experimental Procedures"). This frequency was 4. Sequence Analysis of Mutants-Several of the mutants from donor-directed products produced in the absence of NCp or Klenow products were sequenced (Fig. 4). It is difficult to assign specific errors to RT since the background in the assay was relatively high (see above). Errors may have resulted from RNA polymerase, Taq, or improper insertion. The latter group consisted mostly of plasmids with 2 inserts in which the PvuII cut blunt end of one insert had ligated to a modified EcoRI end of a second insert. The 4 base 5Ј overhang generated by EcoRI was apparently cleaved off by a contaminating nuclease generating a blunt end. This accounted for about one-fourth of the mutated plasmids observed with Klenow (data not shown). The other mutant plasmids from Klenow consisted of single base substitutions and deletions, presumably resulting from errors made by RNA polymerase, or Taq, or Klenow. Some of the observed errors were unique to assays with HIV-RT. Among them were frameshifts within a run of A's (bases 65-68) and multiple substitution mutants at positions 61 and 86. DISCUSSION We have shown that internal strand transfer occurring over a defined region of RNA is highly accurate. This conclusion was based on the error frequency of HIV-RT DNA synthesis products produced by primer extension on a single template, or those which transferred to and were extended on a second acceptor template. Each type of product showed similar levels of errors (Table I). If strand transfer were inaccurate, then transfer products should have more errors than those that had not undergone transfer. Due to the assay background and variability, very small increases in error frequency could not be reliably detected. However, if 2% or more of transfer events were inaccurate this could have been easily detected. A 2% per event error rate for transfer would have increased the error frequency by about 2 ϫ 10 Ϫ4 per base. This calculation is based on 2 additional mutant colonies per 100 total colonies or 2% additional mutants. Since each insert corresponds to 99 synthesized nucleotides (see "Results") the error frequency would be 2/(100 ϫ 99) or about 2 ϫ 10 Ϫ4 . This level of error would increase the 2.8 ϫ 10 Ϫ4 value for full-length products without NCp from Table I (column 2) to 4.8 ϫ 10 Ϫ4 for the transfer products. Thus, if errors occurred in only 1 in 50 transfer events they would have been easily detected. There was some increase in error frequency for transfer products versus donordirected products produced in the presence of NCp. An increase from 2.7 ϫ 10 Ϫ4 to 3.4 ϫ 10 Ϫ4 represents one additional error per about every 14,000 bases copied or a 0.7% per event error rate for transfer. Since this increase is relatively small it may have resulted from normal experimental variation. Several additional experiments would have to be performed to substantiate this difference. What was clear from the results is that from a per base perspective, strand transfer on this particular template was considerably more accurate than RT-directed DNA synthesis. The latter had a maximum error frequency after background subtraction of 2.4 ϫ 10 Ϫ4 in the assay (see "Results"). This corresponds to one error per about 4200 nucleotides. Clearly, errors resulting from strand transfer were well below this value. That does not mean that strand transfer is 100% accurate, but that base misincorporation errors resulting from HIV-RT's infidelity occur at a significantly higher frequency than those resulting from erroneous strand transfer. This may be especially true in vivo where the strand transfer frequency using an spleen necrosis virus based system was estimated to be about 4% per kilobase during a single round of replication (9). If the frequency of an inaccurate strand transfer event was 1%, using this value of 4% to estimate the frequency of recombination per kilobase leads to an error frequency per base of 4 ϫ 10 Ϫ7 (0.01 ϫ 0.04/1000) or 1 error per 2.5 ϫ 10 6 bases on average. Our experiments suggest an error frequency per strand transfer event of less than 1%. If this scenario is reasonable, the contribution of strand transfer to base misincorporations would be essentially negligible.
We should note that the frequency of strand transfer events in our in vitro system is 1-2 orders of magnitude greater than the 4% per kilobase value sited above. We observed frequencies corresponding to average rates of about 1 transfer event per 1000 or 330 nucleotides for reactions in the absence or presence of NCp, respectively (Table I). The high rate observed in our experiments may result in part, from the small sizes of the donor and acceptor templates and the high ratio of acceptor to donor (5:1) (36). The relatively large size of the retroviral genome would likely make alignment of homologous regions more difficult than with the small templates used here. In addition, one might expect that the large number of bases of the genome may increase the likelihood of erroneous transfer events, essentially by increasing the potential targets for transferring DNAs. For example, a nascent DNA could potentially transfer to a region on the acceptor genome hundreds or even thousands of bases away from the region homologous to that used for synthesis of the DNA. Such events are referred to as "nonhomologous" strand transfer since the DNA transfers to a region of a second template different from the region on which it was synthesized. In this type of transfer there is often, but not always, a small region of complementarity between the acceptor RNA and 3Ј terminus of the transferring DNA (37,38). Experiments performed in vivo using a system based on spleen necrosis virus and Moloney murine leukemia virus indicate that nonhomologous recombination occurs at only 1/100th to 1/1000th the frequency of homologous recombination (37). Nonhomologous transfer events may be important for viral transduction (38,39); however, homologous transfer events, being more frequent and less likely to produce defective provirus, presumably contribute more to the genetic diversity of the viral population. The size of the templates used in our experiments limits the potential for evaluating nonhomologous transfer. In fact, any transfers resulting in the deletion or insertion of about 10 or more bases would have been missed by our assays as a result of the way the insert was isolated (by gel purification in which a region of about Ϯ10 bases was excised). Therefore, the assay we used assesses mutations occurring during homologous strand transfer. Our conclusions suggest that such transfers are highly accurate. Others, however, have found strand transfer with HIV-RT to be somewhat inaccurate (26,29). Using an in vitro system designed to test the fidelity of internal transfer within a hypervariable region of the nef gene, Wu et al. (29) reported that strand transfer was more error-prone than RNAdirected DNA synthesis. The overall error frequencies per base for DNA synthesis and strand transfer were 2.8 ϫ 10 Ϫ5 and 6.2 ϫ 10 Ϫ5 , respectively. The value for RNA-directed DNA synthesis is about 10-fold lower than the value calculated from our experiments but is comparable to the values suggested by Boyer et al. (6). Although the error frequency in strand transfer more than doubled in the above experiments, the overall gross increase was 6.2 ϫ 10 Ϫ5 -2.8 ϫ 10 Ϫ5 or 3.4 ϫ 10 Ϫ5 . We note that such an increase would not be reliably detectable in our assays and would represent a tenuous increase over the error frequency which we determined. There were important differences between our experiments and those of Wu et al. (29). Although both used color changes due to loss of ␣-complementation to detect errors, in the experiments of Wu et al. (29) synthesis and transfer occurred over a hypervariable region of the nef gene which was inserted within the multiple cloning site of plasmid pBSM13ϩ, between the ␣-lac promoter and the majority of the coding region for ␣-lac. The insert was in-frame with the downstream ␣-lac gene. Mutations occurring during DNA synthesis or strand transfer within the insert region would disrupt ␣-lac only if the mutations resulted in a termination codon or frameshift. Therefore, the assay was designed to detect mutations resulting in frameshifts only. Since our assay detected both frameshifts and some point mutations (see "Results"), we would likely detect a higher proportion of the total errors. However, since a high proportion of the errors made by HIV-RT on RNA are frameshift errors (6), this still cannot explain the 10-fold difference between the experiments. One very important difference is the sequence of the RNA used in the experiments. We performed synthesis and transfer over the N-terminal region of the ␣-lac gene while Wu et al. (29) used a portion of the HIV genome. It is quite possible that there are significant sequence-dependent differences in error frequencies. This may also explain the increase in errors in strand transfer products. Perhaps there are specific sequences which tend to transfer inaccurately. In support of this Wu et al. (29,40) noted that although transfer occurred from several regions of the template, a particular run of 4 C's had a high error frequency only after strand transfer. Perhaps our template did not contain a hypermutable transfer sequence. Finally, it is possible that strand transfer on our template occurred essentially from a single location that just happened to promote high fidelity transfer. However, as was noted under "Results," variations in pause site intensities in the absence or presence of acceptor suggested that transfer occurred from several locations on the template, although the relative contribution of each site to transfer was not assessed.
Results from others have suggested that strand transfer of minus strand strong stop DNA may be inaccurate due to nontemplate-directed base insertions at the 3Ј terminus of the transferring DNA (26). Mutations of this type would have resulted in frameshifts in our assay system and were not detected. Perhaps the nature of strand transfer events occurring from template termini (i.e. strong stop transfers) differs from those occurring from within internal regions of a template.
It was interesting that the error frequency calculated in our experiments was very close to the 1 error per 6900 base value determined by Ji and Loeb (8). When the error rate for Klenow was used as background we calculated a minimum error frequency of 1 per 7700 bases (see "Results"). Although the approach used by Ji and Loeb (8) differed from ours, the sequence and length of RNA used in the two studies were comparable. Hence it is not surprising that similar values were obtained. Many of the mutations observed by Ji and Loeb (8) were also observed in our experiments, specifically frameshifts within a particular run of adenosines (bases 65-68 in Fig. 4). There were also differences, particularly in the observation of guanosine to thymidine transversions of which we detected only one (base 64). In contrast, a significantly lower error frequency for RNA-directed DNA synthesis, especially with respect to base substitutions, was observed using a larger region of ␣-lac that contained the sequences used in our experiments (6). We, like Ji and Loeb (8), observed a substantially higher proportion and number of base substitutions in comparison to frameshifts while fewer substitutions were observed by Boyer et al. (6). We note that the methodologies used in the three studies were considerably different. Our strategy required several enzymatic steps and a ligation to score both transfer and donor-directed products, while Boyer et al. (6) employed the least manipulation in their approach. Due to the many steps, the background in our experiments was relatively high, perhaps accounting for up to one-half of the total errors (see "Results"). This made it difficult to be sure that a particular error was due to RT and not one of the other enzymes used in the process. However, in comparing Klenow and RT products, some errors were observed multiple times and only with RT, implying that they resulted from this enzyme (Fig. 4). All things considered, the lower background of the assays used by Boyer et al. (6) and Ji and Loeb (8) make these approaches more informative with respect to analyzing specific errors made by HIV-RT.
In conclusion, our work indicated that strand transfer on the template used in these experiments was highly accurate, while experiments using a different template (29) showed that some transfer events are error prone. Taken together the results imply that the fidelity of strand transfer may be highly sequence dependent.