The Nature of Human Immunodeficiency Virus Type 1 Strand Transfers*

The diploid nature of human immunodeficiency virus type 1 (HIV-1) suggests that recombination serves a central function in virus replication and evolution. A system was developed to examine HIV-1 strand transfers, including the obligatory DNA primer strand transfers as well as recombinational crossovers during reverse transcription. Sequence heterogeneity between different strains of HIV-1 was exploited for examining primer transfer events. Both intra- and intermolecular primer transfers were observed at similar frequencies during minus-strand DNA synthesis, whereas primer transfers during plus-strand DNA synthesis were primarily intramolecular. Sequence analysis of long terminal repeats from progeny proviruses also revealed a high rate of homologous recombination during minus-strand synthesis, corresponding to an overall rate of approximately three crossovers per HIV-1 genome per cycle of replication. These results imply that both viral genomic RNAs serve as templates during HIV-1 reverse transcription and that primer strand transfers and recombination may contribute substantially to the rapid genetic variation of HIV-1.

the encapsidation of a second RNA genome, and/or that recombination is a potent force in viral evolution, requiring the second genome to serve as a recombination template (2).
DNA primer strand transfers have been investigated for the avian oncoretrovirus, spleen necrosis virus (SNV) (3)(4)(5). It was found that the plus-strand DNA primer transfer is exclusively intramolecular, and the most recent study indicates that the minus-strand DNA primer transfer is primarily intramolecular except for recombinant proviruses, which can undergo either intra-or intermolecular primer strand transfers (3)(4)(5). With the emergence of complex retroviruses in recent years, it was not known whether the previous studies investigating primer transfer events using the simpler avian oncoretrovirus, SNV, apply to complex retroviruses such as HIV-1.
Recombination has been identified as an important means to incorporate genetic changes into retrovirus genomes. The rate of recombination for SNV was found to be 4 ϫ 10 Ϫ5 per bp per replication cycle (4,5). Although the rate of HIV-1 recombination was not examined, previous phylogenetic analyses indicate that 5-10% of sequenced HIV-1 strains are recombinant containing genetic material from different subtypes (6). Further information regarding the rate and mechanisms of HIV-1 recombination are needed for a more thorough understanding of HIV-1 recombination and its role in HIV-1 variation.
In this report, we present a study examining both recombination and the obligatory HIV-1 primer stand transfers. The study was carried out using two vectors based on different strains of HIV-1. The vectors were constructed with minimal deletions such that the size of the vectors was similar to wildtype HIV-1 genomic RNA. After a single cycle of vector virus replication, the nature of primer strand transfers between viral sequences was evaluated. The results indicate that HIV-1 minus-strand DNA primer transfers can be either intra-or intermolecular and that plus-strand primer transfers are intramolecular. A high frequency of homologous recombination was also observed during minus-strand U3 synthesis, which if extrapolated to the entire genome suggests that, on average, approximately three crossovers occur during each round of HIV-1 replication. Possible recombination events during plusstrand primer synthesis were also observed at a lower frequency. These results suggest that both RNAs are typically used during HIV-1 reverse transcription and that intermolecular strand transfers and homologous recombination can contribute significantly to HIV-1 genetic variation.

EXPERIMENTAL PROCEDURES
Plasmid Construction-pHIV-gpt HXB2 and pSG3.1 were obtained from the AIDS Repository (7,8). In HIV-gpt HXB2 , part of env was replaced with SV-gpt (Escherichia coli xanthine-guanine phosphoribosyltransferase (gpt) gene under control of the SV40 promoter). To construct the SG3puro BCSG3 vector, a 1.1-kb deletion was made first in pSG3.1 in the provirus envelope gene from the Bst1107 I site at HIV-1 * This work was supported by National Institutes of Health Grants CA50777 and AI34834 and by an award from the Milstein Family Foundation. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18  BCSG3 nucleotide 5965 to the BsrG I site at nucleotide 7086. The deletion was replaced with a 1.4-kbp EcoRI to BsmI fragment from pHY18puro, which contains the SNV U3 promoter and puromycin resistance gene (puro) coding sequence (map available upon request).
Cell Culture-A human embryonal kidney cell line, 293, was grown in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum. 69TIRevEnv cells were grown in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum, 0.2 mg of G418 per ml, 0.1 mg of hygromycin per ml, and 2 g of tetracycline per ml (9). The HeLaT4 cell line was a gift from Michael Emerman and was grown in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum and 0.1 mg of hygromycin per ml.
Transfection and Infection-Transfections were performed using the modified calcium phosphate precipitation method (10). For infections, 2 ϫ 10 5 target cells in 60-mm-diameter dishes were first treated with 2 g of polybrene per ml for 30 min. Treated cells were subsequently inoculated with virus in 0.3 ml of medium. Vector virus titers were determined by infection of HeLaT4 cells with 10-fold serial dilutions of the virus stocks. At 24 h after infection, the cells were selected for resistance either to GPT (250 g/ml xanthine, 15 g/ml hypoxanthine, and 7 g/ml mycophenolic acid), to determine HIV-gpt HXB2 vector virus titers, or to puromycin (1 g/ml), to determine SG3puro BCSG3 vector virus titers.
Analysis of Proviral DNA-Proviral DNA was amplified from genomic DNA of infected cell clones by the polymerase chain reaction (PCR) using Taq DNA polymerase (Promega). The primer pair used to amplify the 5Ј LTR was 5Ј-CTACCACACACAAGGCTACT-3Ј, located close to the 5Ј end of U3, and 5Ј-CTCGCACCCATCTCTCTCCTT-3Ј, located in the 5Ј end of the gag gene. The primer pair used to amplify the 3Ј LTR was 5Ј-CAAGAGGAGGAGGAGGTGGGT-3Ј, located in the nef gene outside of the 3Ј LTR, and 5Ј-CTAGAGATTTTCCACACT-GACT-3Ј, at the 3Ј end of U5. Amplified proviral DNA was purified using QIAquick purification columns (QIAGEN Inc.), digested with AvaII or DdeI, and separated through a 3.5% NuSieve agarose gel (FMC, Rockland, ME). To sequence the PCR products, amplified products were separated from the primers through a 1% agarose gel and were subsequently purified using QIAquick gel extraction kit (QIA-GEN). The purified DNA was cloned into the pGEM-T vector using pGEM-T Easy vector system according to the manufacturer's direction (Promega). The cloned DNA was sequenced by automated DNA sequencing using either a T7-or SP6-specific primer.
Calculation of the Composition of Virion RNA-Because intermolecular primer strand transfers cannot be recognized from homodimeric virions, the number of proviruses assayed must be corrected to exclude proviruses from homodimeric virions. The anticipated fraction of the progeny proviruses arising from heterozygous virions was determined using the Hardy-Weinberg equation (11), G 2 ϩ 2GP ϩ P 2 ϭ 1, where G ϭ GPT titer/(GPT titer ϩ PURO titer) and P ϭ PURO titer/[GPT titer ϩPURO titer]. G 2 represents the fraction of homozygous HIV-gpt HXB2 virions, 2GP the fraction of heterozygous virions, and P 2 the fraction of homozygous SG3puro BCSG3 virions. The average titer was 4.5 ϫ 10 4 colony-forming units (cfu)/ml for HIV-gpt HXB2 , and 5.6 ϫ 10 2 cfu/ml for SG3puro BCSG3 . Therefore, G ϭ 0.99 and P ϭ 0.01. The percentage of heterozygous virions among all virions containing SG3puro BCSG3 RNA was 99% (GP/(P 2 ϩ GP) ϭ 99%). Thus, of the 86 puro r progeny cell clones examined, proviruses from 85 clones should have resulted from heterozygous virions.
Fisher's exact test was used to calculate whether there was a significant difference in minus-strand primer transfer pattern between the two progeny groups that had minus-strand primer initiated from either HIV-gpt HXB2 or SG3puro BCSG3 . p Ͼ 0.05 was considered insignificant.

Protocol for the Study of Primer Strand Transfers in a Single
Cycle of Virus Replication-Primer strand transfer events were studied by examining the proviral progeny from heterozygous virions after a single cycle of virus replication (3,4). Two different HIV-1 vectors were used to generate heterozygous vector virions. HIV-gpt HXB2 and SG3puro BCSG3 are env-defective HIV-1 vectors derived from different HIV-1 strains, HXB2 and BCSG3, respectively ( Fig. 2A) (7). The sequences of the two strains differ by approximately 5%, providing appropriate heterogeneity in the viral LTRs for examining the nature of primer strand transfers. The majority of the vector sequences are viral (86ϳ89%), with only a deletion of part of env, leaving the rest of the viral genome intact. Therefore, the vector RNAs produced are similar in size to the authentic HIV-1 genome (9.2 ϳ 9.6 kb).
The protocol employed to study the primer strand transfer is outlined in Fig. 2B. To introduce single copies of HIV-gpt HXB2 and SG3puro BCSG3 proviruses into the HIV-1 packaging cell line 69TIRevEnv (9), each vector was first produced as a pseudotyped virus by cotransfection with an amphotropic murine leukemia virus env expression vector pEnvAm (12) into 293 cells. Pseudotyped HIV-gpt HXB2 vector virus was then used to inoculate 69TIRevEnv cells at a low multiplicity of infection followed by selection for gpt. Individual gpt r cell clones were isolated and expanded, and Southern blotting was used to confirm that each cell clone harbored a single copy of the HIV-gpt HXB2 provirus (data not shown). These cell clones were subsequently inoculated with pseudotyped SG3puro BCSG3 vector virus at low multiplicity of infection, followed by selection with puromycin. After puromycin selection, individual cell clones were isolated, and Southern blotting was again employed to confirm that each cell clone contained a single copy of the SG3puro BCSG3 vector provirus (data not shown). Eight independent producer cell clones containing single copies of both HIV-gpt HXB2 and SG3puro BCSG3 proviruses were established. These clones were genetically stable and did not produce infectious vector virus when maintained in the presence of tetracycline (9). To initiate a single cycle of replication, tetracycline was removed, and vector virus released from the producer cells was used to inoculate CD4-positive HeLaT4 cells. Infected He- LaT4 cells were selected with puromycin, and 86 individual puromycin-resistant (puro r ) clones were expanded for proviral DNA analysis. In this system, the vector virus is restricted to a single cycle of replication because the producer cells are CD4-negative and cannot be reinfected by vector virus, whereas the HeLaT4 target cells lack HIV-1 Env and thus fail to produce new infectious particles.
Cells harboring two vector proviruses will produce homozy-

FIG. 2. Vectors and protocol utilized to examine primer strand transfers in a single cycle of replication. A,
HIV-1-based retroviral vectors for studying primer strand transfers. HIV-gpt HXB2 is based upon HIV-1 strain HXB2 (7). SG3puro BCSG3 is based upon HIV-1 strain BCSG3 (8). SV-gpt represents SV-40 early promoter and gpt. SNV-puro represents spleen necrosis virus U3 promoter and puro. B, protocol for the study of primer strand transfer events. HIVgpt HXB2 and SG3puro BCSG3 were cotransfected with pEnvAm, a plasmid expressing amphotropic murine leukemia virus env, into parallel cultures of 293 cells (12). Pseudotyped HIV-gpt HXB2 vector virus was used to infect a CD4-negative HIV-1 Env-inducible cell line, 69TIRevEnv (9). Individual cell clones containing a single copy of HIV-gpt HXB2 provirus were established. Pseudotyped SG3puro BCSG3 was subsequently used to infect the established 69TIRevEnv cell clones containing HIV-gpt HXB2 . Individual producer cell clones containing a single copy of both SG3puro BCSG3 and HIV-gpt HXB2 proviruses were established. Upon induction, vector virus was harvested from the producer cell clones and was used to infect CD4-positive HeLaT4 target cells, followed by selection and isolation of target cell clones. The proviruses integrated in these target cells were further analyzed. Replication from a provirus in a producer cell to a provirus in a target cell comprised a single cycle of virus replication. gous virions containing two identical viral RNAs as well as heterozygous virions containing two different viral RNAs. Only progeny derived from heterozygous virions are informative for the analysis of primer strand transfers. Among the eight producer cell clones studied, the average vector virus titer for SG3puro BCSG3 was about 100 times lower than that for HIVgpt HXB2 . Thus, according to the Hardy-Weinberg equation (11), the large majority of proviruses (ϳ99%) in transduced HeLaT4 cell clones selected for puromycin resistance should result from infection by heterozygous virions (see "Experimental Procedures"). This prediction was experimentally borne out (see below).
HIV-1 Minus-strand Primer Transfers Can Be either Intramolecular or Intermolecular-Proviruses in puro r target cell clones were analyzed for primer strand transfers using PCR in combination with restriction enzyme mapping and DNA sequencing. As described previously (3), one can determine whether intra-or intermolecular primer strand transfers occur during reverse transcription by analyzing the LTRs of progeny proviruses derived from heterozygous virions. When both U3 and U5 sequences originate from the same vector, the minus-strand primer transfer is intramolecular. When U3 and U5 sequences originate from different vectors, the minus-strand primer transfer is intermolecular. The sequence differences between HIVgpt HXB2 and SG3puro BCSG3 in U3 and U5 can be identified by restriction mapping (Fig. 3). Digestion with AvaII reveals U3 sequence differences, whereas digestion with DdeI indicates U5 sequence differences between the two vectors (Fig. 3). By determining the origin of the U3 and U5 sequences, the nature of the minus-strand primer transfer event can be deduced.
To analyze the minus-strand primer transfers, the 3Ј LTRs were amplified from 86 independent puro r target cell clones using primers from the nef region outside the LTR and the 3Ј end of U5 (Fig. 3). The resulting PCR products were separately digested with AvaII and DdeI. Representative digestions for five clones are depicted in Fig. 3. In the DdeI digestion, a 488-bp fragment indicates that the 5Ј upstream region of the 3Ј LTR is of BCSG3 origin, whereas a 444-bp fragment indicates that this region is of HXB2 origin. In addition, a 155-bp fragment indicates that U5 is of BCSG3 origin, whereas a 121-bp fragment indicates that U5 is of HXB2 origin. For the AvaII digestion, three fragments (345, 237, and 158 bp) indicate that U3 is of BCSG3 origin, whereas an uncut 741-bp fragment indicates that U3 is of HXB2 origin. The DdeI digestion of the 3Ј LTR of clone 1 produced a 444-bp fragment and a 121-bp fragment. Therefore, for clone 1, both the 5Ј upstream region and the U5 are of HXB2 origin. The AvaII digestion of clone 1 produced an uncut 741-bp fragment, indicating that the U3 is of HXB2 origin. Thus, clone 1 had undergone an intramolecular minus-strand transfer with both U3 and U5 originating from HIV-gpt HXB2 . A similar analysis is shown for four additional clones, producing the pattern depicted under each gel lane. Clone 3 had an intramolecular minus-strand transfer with both U3 and U5 originating from SG3puro BCSG3 . Clones 2 and 4 had intermolecular minus-strand transfers with U5 sequences originating from HIV-gpt HXB2 and U3 sequences originating from SG3puro BCSG3 . Clone 5 had an intermolecular strand transfer with U5 originating from SG3puro BCSG3 and U3 originating from HIV-gpt HXB2 .
Of the 86 progeny target cell clones analyzed, 35 clones showed intramolecular and 40 clones showed intermolecular minus-strand primer transfer patterns. Sequencing results from 10 randomly picked clones were consistent with the restriction enzyme mapping (data not shown), indicating the digestion patterns reflected true primer strand transfer events rather than PCR errors or mutation at the restriction enzyme sites. The distribution of intra-versus intermolecular strand transfers among progeny derived from different producer cell clones was similar ( Table I). 11 of the 86 clones yielded restriction patterns that were not consistent with simple intra-or intermolecular strand transfer events. These clones were further analyzed by DNA sequencing and are described in more detail in a later section.
Consistent with the prediction of Hardy-Weinberg equation (see "Experimental Procedures"), it was found that all 86 proviruses were derived from heterozygous virions. Based on analysis of the LTRs, 73 of the 86 clones were clearly products of heterozygous virions; 22 had HIV-gpt HXB2 LTRs but were puro r , 40 had undergone intermolecular minus-strand DNA primer transfer, and 11 had undergone interstrain recombination during U3 synthesis (see below). The remaining 13 clones had SG3puro BCSG3 LTRs, but all of these clones have undergone homologous recombination with HIV-gpt HXB2 sequences elsewhere in the genome. 2  analysis was not affected by progeny from homozygous virions.
HIV-1 Plus-strand Primer Transfers Are Intramolecular-The nature of the plus-strand primer transfer event can be determined by comparing the restriction enzyme digestion patterns of the 5Ј and 3Ј LTRs (Fig. 4). Two identical LTRs would indicate that the plus-strand primer transfer was intramolecular. Two different LTRs would suggest that plus-strand primer transfer was intermolecular or that recombination and/or mutation had occurred. The sequence origin of U3 and U5 in the 5Ј LTR was determined in a fashion similar to that described above.
The 5Ј LTRs of the 86 progeny cell clones were subjected to PCR using a primer close to the 5Ј end of U3 sequence and a primer at the 5Ј end of the gag gene. Similar to the 3Ј LTR analysis, the PCR products were separately digested with DdeI and AvaII. Representative digestions for five clones are depicted in Fig. 4. In the DdeI digestion, a 323-bp fragment indicates that the U5 is of BCSG3 origin, whereas two fragments (201 and 121 bp) indicate that U5 is of HXB2 origin. For the AvaII digestion, three fragments (342, 255, and 150 bp) indicate that U3 and the region immediately 3Ј to the LTR are of BCSG3 origin, whereas an uncut 747-bp fragment indicates that the U3 and 3Ј downstream regions are of HXB2 origin. For the AvaII digestion, two other patterns are possible. Two fragments, 405 and 342 bp, indicate that U3 is of BCSG3 origin and the 3Ј downstream region is of HXB2 origin. Two fragments, 597 and 150 bp, indicate that U3 is of HXB2 origin and the 3Ј downstream region is of BCSG3 origin. The DdeI digestion of the 5Ј LTR of clone 1 produced a 201-bp fragment and a 121-bp fragment. Therefore, for clone 1, U5 is of HXB2 origin. The AvaII digestion of the 5Ј LTR of clone 1 produced an uncut 747-bp fragment. Therefore, for clone 1, U3 and the 3Ј downstream region are of HXB2 origin. A similar analysis for four additional clones is shown, producing the pattern depicted at the bottom of Fig. 4. Thus, the U3 and U5 sequences for clone 1 originated from HIV-gpt HXB2 and for clone 3 originated from SG3puro BCSG3 . Clones 2 and 4 had U5 sequences from HIVgpt HXB2 and U3 sequences from SG3puro BCSG3 . Clone 5 had U5 sequences originating from SG3puro BCSG3 and U3 sequences from HIV-gpt HXB2 . As compared with the 3Ј LTR sequences analyzed in Fig. 3, the U3 and U5 origin patterns were identical for both LTRs among the five clones, indicating intramolecular primer strand transfers during plus-strand synthesis.
Of 86 progeny target cell clones, 5Ј LTR sequences were successfully amplified from 84 proviruses. Sequence changes might have occurred that affected PCR amplification in 2 of 86 clones. 81 clones displayed intramolecular primer strand transfer during plus-strand DNA synthesis (Table I). 13 of 81 clones were selected randomly and subjected to DNA sequencing. The results were consistent with the restriction enzyme mapping (data not shown). The PCR product of one of the 84 clones was consistently smaller in size than the rest of the clones in three separate experiments, effectively ruling out a PCR artifact. Sequencing analysis indicated a 400-bp deletion, encompassing the majority of the U3 and part of the R sequences in the 5Ј LTR. The U5 region was intact and was from SG3puro BCSG3 . The U3 remnant was compatible with the SG3puro BCSG3 U3 sequence. Despite the deletion, the U3 and U5 origins matched the 3Ј LTR, suggesting this clone had also undergone intramolecular plus-strand primer transfer.
Two clones yielded differences between the 5Ј LTR and 3Ј LTR. In clone 158, U5 sequences were from HIV-gpt HXB2 in both 5Ј and 3Ј LTRs, whereas U3 was from HIV-gpt HXB2 in the 5Ј LTR and from SG3puro BCSG3 in the 3Ј LTR. In clone D3, U3 sequences originated from HIV-gpt HXB2 in both 5Ј and 3Ј LTRs, but U5 was from SG3puro BCSG3 in the 5Ј LTR and from HIVgpt HXB2 in the 3Ј LTR. These would not be the expected patterns from intermolecular primer strand transfers during plusstrand synthesis. Instead, it is more likely that they resulted  from recombination during plus-strand primer synthesis. The growing point of plus-strand primer synthesis might have crossed over to the other minus-strand DNA template. Alternatively, recombination might have occurred between the two plus-strand primer DNAs.
Homologous Recombination Events Occur during U3 Synthesis-As noted above, 11 of 86 clones yielded unexpected digestion patterns within the 3Ј LTR, inconsistent with either a simple intra-or intermolecular minus-strand DNA primer transfer. The 11 clones were further examined by DNA sequencing of both 5Ј and 3Ј LTRs. Sequence analysis indicated that 7 of those clones (clones 106, 113, 151, 153, D1, D17, and D23) had undergone an intramolecular minus-strand primer transfer and one homologous recombination event (a single crossover) (Fig. 5). Four of the clones (clones 141, 143, 155, and D30) had undergone an intermolecular transfer during minusstrand primer synthesis plus homologous recombination, in which three had undergone a single crossover and one a double crossover (Fig. 5). The 5Ј LTRs of the 11 clones showed recombination patterns consistent with those observed in the 3Ј LTRs, suggesting that the recombination events occurred during minus-strand DNA synthesis. Given the target size of the FIG. 5. Recombination during minus-strand U3 synthesis. A, sequence comparison of the 3Ј LTR between HIV-gpt HXB2 and SG3puro BCSG3 and indication of the sections where recombination occurred during U3 synthesis in the recombinant clones. The sequences of the HIV-gpt HXB2 and SG3puro BCSG3 3Ј LTRs as well as some sequence 5Ј to the LTR were compared using GCG Bestfit program (Genetics Computer Group, Madison, WI). U3, R, and U5 are indicated. Absence of a vertical line indicates sequence differences between HIV-gpt HXB2 and SG3puro BCSG3 . The sequences of the two strains differ by 4.5% in the 3Ј LTR. Sequencing analysis of the 11 recombinant clones revealed the crossover regions, which are denoted with dotted lines. The cell clone that underwent recombination in a particular region is indicated above the dotted line. H represents HIV-gpt HXB2 sequence, and S represents SG3puro BCSG3 sequence. A dotted line starting with H at the 3Ј side and ending with S at the 5Ј side indicates that during U3 synthesis, the growing strand originated from HIV-gpt HXB2 sequence and crossed over to SG3puro BCSG3 sequence. On the other hand, a dotted line starting with S at the 3Ј side and ending with H at the 5Ј side indicates that during U3 synthesis, the growing strand originated from SG3puro BCSG3 and crossed over to HIV-gpt HXB2 sequence. The "A" underneath nucleotide 159 denotes a G to A mutation in the crossover region of clone D30. B, summary of recombinant cell clones. Five different classes of recombinants are illustrated; the numbers to the right indicate how many of each different type occurred. 1 , includes cell clones 113, 151, 153, D1, and D17; 2 , includes cell clones 106 and D23; 3 , includes cell clone 143; 4 , includes cell clones 141 and 155; and 5 , includes cell clone D30.
U3 region (452 bp), the recombination rate during minusstrand DNA synthesis was 3 ϫ 10 Ϫ4 per bp per replication cycle. As indicated in Fig. 5A, homologous recombination events can occur throughout U3. However, four independent proviruses generated from two different producer cell clones underwent a homologous recombination in the same segment ( nucleotides  192 ϳ 217, clones 141, 151, 153, and D1, Fig. 5A), suggesting a potential hot spot for recombination. One mutation, a G to A mutation in clone D30 (nucleotide 159, Fig. 5A), was observed in the crossover regions among the 11 recombinant clones. DISCUSSION We have examined the nature of HIV-1 primer strand transfers and have also gained insight into the rate and the strandedness of recombination. Both intra-and intermolecular primer strand transfers were observed during minus-strand DNA synthesis, whereas only intramolecular strand transfers occurred during plus-strand DNA synthesis. Sequence analysis of the LTRs also revealed a high rate of homologous recombination, which if extrapolated to the entire genome, indicates crossovers occur at a rate of almost three per genome per replication cycle during minus-strand DNA synthesis.
The Nature of HIV-1 Primer Strand Transfers-Both intraand intermolecular strand transfers were observed during minus-strand DNA synthesis, and they occurred with similar frequencies ( Table I). The pattern of minus-strand primer transfer in HIV-1 seems to contrast with that reported for SNV (5). It was observed that minus-strand primer transfers were mainly intramolecular for SNV (5), suggesting that this aspect of HIV-1 replication differs from SNV. This might be related to the significantly higher rate of homologous recombination observed for HIV-1 (3 ϫ 10 Ϫ4 /bp/cycle) compared with SNV (4 ϫ 10 Ϫ5 /bp/cycle) (4,5). In the nonrecombinant SNV proviruses, minus-strand DNA primer transfers were primarily intramolecular (95 out of 96 nonrecombinant proviruses) (5). However, among the smaller population of recombinant SNV proviruses (10%), similar numbers of intra-and intermolecular minusstrand primer transfers were observed (4,5). Because of the apparently higher rate of HIV-1 recombination, the majority of the HIV-1 proviruses are presumably recombinant, maintaining the correlation between disordered minus-strand primer transfer and recombination. Thus, the underlying significant difference between HIV-1 and SNV may lie with differences in their rates of recombination.
Differences in the experimental approach may also contribute to the contrasting results obtained for HIV-1 and SNV (5). In this study, two vectors similar in size to wild-type HIV-1 were used, the majority of sequences in the vectors were viral, and natural viral heterogeneity between the two HIV-1 strains was employed to analyze primer strand transfers. These features differed from previous studies (5,13). However, it is most likely that the contrasting results are due to inherent differences between the two viruses. Three potential differences between the viruses are described below. First, the biochemical properties of the RTs might differ such that HIV-1 RT promotes more frequent strand transfers. Second, the HIV-1 accessory proteins, which have been shown to effect reverse transcription (14 -16), might also influence the rate and nature of strand transfers. Third, the structure of the HIV-1 core particle, which is clearly different from that of oncoretroviruses such as SNV (17), might place less constraints upon HIV-1 strand transfers. At this juncture, it is unclear which, if any, of these factors accounts for the differences. However, all three possible explanations provide direction for future studies of the underlying mechanism of retroviral strand transfers.
Also noteworthy is that there was about a 100-fold difference in titer between the two vectors used in this study. Moreover, minus-strand primer synthesis initiated more efficiently from HIV-gpt HXB2 than from SG3puro BCSG3 genomic RNA. The ratio was about 3:1 (66 proviruses initiated from HIV-gpt HXB2 and 20 from SG3puro BCSG3 , data not shown). As demonstrated in previous reports, different HIV-1 viral isolates may exhibit different replication characteristics (18,19). Aside from the differences observed in this study, both HIV-gpt HXB2 and SG3puro BCSG3 genomes successfully served as templates for strand transfers.
Concerns may arise that the strand transfers examined were intrinsically linked to recombination due to the experimental procedure and that proviruses which had undergone recombination between the marker gene (puro) and the 3Ј LTR may not accurately reflect the quality of the primer strand transfer. Actually, the proviruses with U3s of SG3puro BCSG3 origin did not have to undergo recombination to be scored. Within this subpopulation, there are two groups: those proviruses with U3 of SG3puro BCSG3 (S) origin and U5 of HIV-gpt HXB2 (H) origin, and those proviruses with both U3 and U5 of S origin. The proviruses with the U3/U5 pattern SH were products of intermolecular minus-strand transfer, whereas those with the pattern SS were products of intramolecular minus-strand transfer. Among the 86 proviruses analyzed, it has been ascertained that 28 SH clones and 10 SS clones did not undergo recombination between the marker and the 3Ј LTR representing a population not intrinsically linked to recombination. 3 Since reverse transcription initiated three times more efficiently from HIV-gpt HXB2 , producing three times more clones with U5 of H origin, the relative ratio of intermolecular (SH, 28 clones) to intramolecular (SS, 10 clones) can be normalized by multiplying the SS group by three. Thus, within the population for which selection was not linked to recombination, the relative number of inter-to intramolecular strand transfers was equivalent indicating that the experimental approach employed accurately reflects the nature of the primer strand transfers.
HIV-1 Recombines at a High Rate during Minus-strand DNA Synthesis-11 clones underwent recombination during minusstrand U3 synthesis, suggesting a recombination rate of approximately 3 ϫ 10 Ϫ4 per bp per replication cycle during minus-strand DNA synthesis, or about three crossovers per genome per replication cycle. The high rate of recombination in the relatively small U3 region (452 bp) is consistent with the observation that the U3 of a high proportion of puro r proviruses has originated from HIV-gpt HXB2 (35 of 86), so they would have undergone at least one recombination in the 1.9-kb region between the 3Ј LTR and puro to retain puro marker gene. This homologous recombination rate of HIV-1 is 5-15-fold higher than that reported for SNV (4,5), suggesting that different retroviruses have different recombination rates.
Three models have been proposed to explain retroviral recombination (20 -23). Both the "forced copy choice" model and the "minus-strand exchange" model favor recombination during minus-strand DNA synthesis. The "forced copy choice" model hypothesizes that reverse transcriptase switches template when it encounters RNA breaks, whereas the "minusstrand exchange" model suggests that low processivity of RT results in crossovers during minus-strand synthesis without the need of a damaged template (1,20,22). The "strand displacement assimilation" model proposes that DNA fragments are displaced during plus-strand DNA synthesis from one template and subsequently assimilated by the plus-strand DNA synthesized from the other template, causing recombination during plus-strand synthesis (21, 23). This study suggests that homologous recombination occurs most frequently during mi-nus-strand DNA synthesis because the majority of the crossovers were found in both LTRs (Figs. 1 and 5B). Although these experiments do not distinguish between the "forced copy choice" and "minus-strand exchange" models, given the high rate of recombination and the reported low processivity of HIV-1 RT (24, 25), we would favor the "minus-strand exchange" model. The limited number of clones that apparently underwent recombination during plus-strand synthesis suggests that "strand displacement assimilation" also occurs but at a lower frequency. However, because of the relatively small number of plus-strand recombinants observed, further analysis is required to more rigorously evaluate the relative contributions of minusstrand and plus-strand crossovers to HIV-1 recombination.
12 crossover segments were identified by DNA sequencing. 8 were spread relatively randomly throughout U3, whereas 4 of 12 were found within a 24-bp segment, suggesting that this represents a relative hot spot for recombination (Fig. 5A). It is possible that the primary sequence and/or RNA secondary structure in this segment induces RT to pause, providing a greater opportunity for RT to switch strands. This is supported by in vitro experiments demonstrating that pause sites enhance strand transfers by HIV-1 RT (26). However, one characteristic of HIV-1 RT-mediated strand transfers observed in vitro does not seem to occur during replication, namely the correlation between the very high frequency of mutation and strand transfers. It was previously reported using in vitro assays that HIV-1 RT incorporated additional bases about 50% of the time beyond the 5Ј end of both RNA and DNA template ends (27,28), and it was shown that mutations occurred 30% of the time at one frequently utilized strand transfer point (26). It was suggested that if this occurs in vivo, it would contribute significantly to HIV-1 genetic variation (27,28). In the 12 crossover segments identified, only one mutation was observed, yielding a G to A mutation (Fig. 5A). Whether this mutation occurred during recombination or is a random error of RT is not known. Also, from the 10 clones subjected to sequencing analysis in the 3Ј LTR, no mutation was observed at the U3 and R border, where such mutation would be expected upon formation of the blunt end before transfer. Thus, the DNA strand transfers in vivo did not seem to increase the rate of mutagenesis to the level predicted by the in vitro experiments. Similarly, in an in vivo assay, Zhang and Temin (29) reported that recombination is not an error-prone process for Moloney murine leukemia virus. The difference between in vivo and in vitro studies suggests that viral factors and/or cellular components or conditions can affect the mechanism of HIV-1 recombination.
HIV-1 Reverse Transcriptase Likely Utilizes Both Virion RNAs, Relevance for Generation of Genetic Diversity-We observed that intermolecular minus-strand primer transfers occur about half the time and that there is a high rate of homologous recombination. These data suggest that both virion RNAs are typically used during HIV-1 replication to form a single provirus. This seems to contrast somewhat with a previous study using SNV, where it was concluded that one RNA is sufficient for retrovirus replication (5). Although we do not exclude the possibility that one RNA molecule is sufficient for HIV-1 replication, these results indicate that it is the exception rather than the rule. The higher recombination rate intrinsic to HIV-1 would seem to be responsible for this difference.
Is genetic exchange between viral genomes of heterodimeric virions an important influence upon HIV-1 variation? This mechanism of genetic variation requires coinfection of the same cell by heterogeneous virions. Although the frequency of coinfection in vivo is unknown, recent studies on various HIV-1 isolates suggest that coinfection in vivo may not be a rare phenomenon (6, 30 -34). Given the large virus number (10 10 virions produced per day), rapid turnover rate of HIV-1 (t 1/2 ϭ 0.24 days) in vivo (35)(36)(37), and potentially higher local viral concentration in lymphoid organs, the percentage of mosaic virus generated through intermolecular strand transfers and recombination might be fairly high. From an evolutionary point of view, utilization of both viral genomic RNAs is in the best interest of retroviral replication. It provides the virus with maximum genetic flexibility and reduces the chance of leading to a replicative "dead end" (2).
The results presented in this study reveal the nature of HIV-1 primer strand transfer events and the recombination rate of HIV-1 during minus-strand DNA synthesis. These results should not only help to further the understanding of the HIV-1 replication process but also have an impact on drug and vaccine development and assessment.