The kissing hairpin sequence promotes recombination within the HIV-I 5' leader region.

The role of RNA-RNA template interactions in facilitating recombination during reverse transcription of minus strand DNA has been examined. The tested hypothesis is that template switching by reverse transcriptase is promoted at sites where homologous regions of two RNAs are brought in close proximity via stable intertemplate interactions. Frequency and distribution of template switching between homologous donor and acceptor RNAs were examined within the human immunodeficiency virus type I (HIV-I) 5'-untranslated region (UTR) containing the dimer initiation sequence (DIS). Results were compared with control nondimerizing templates from the pol region. The dimerizing UTR templates displayed a 4-fold higher transfer efficiency than the control. A striking 53% of transfers in the UTR mapped near the DIS, of which two-thirds occurred immediately 5' to this sequence. In the UTR template, deletion of the DIS hairpin disrupted template dimerization and caused a 4-fold drop in transfer efficiency. Insertion of the DIS within the pol template increased both dimerization and transfer efficiency. Transfer distributions revealed that in both sets of templates, DIS-induced dimerization increased the efficiency of transfers across the whole template, with the transfers peaking around the dimerization site. Overall, these results suggest that template dimerization facilitated by the unique geometry of the DIS-promoted kissing interactions effectively promotes recombination within the HIV-I 5'-UTR.

Retroviruses are characterized by a diploid, positive sense RNA genome. After entry into the host cell, the viral reverse transcriptase (RT) 1 copies it into double-stranded DNA by a process utilizing both DNA polymerase and RNase H activities. Completion of synthesis of both minus and plus strand DNA involves template switching events in which the strong stop DNA is transferred from the original template to a complementary sequence at the 3Ј end of either the same or a second homologous copy of the template (1). In addition to these obligatory transfers, strand transfers involving internal regions of the RNA or complementary DNA have also been demonstrated to occur during replication (2)(3)(4). These include both intrastrand (resulting in deletions or duplications) and interstrand transfers (resulting in recombination).
The presence of two RNA copies provides a functional advantage during reverse transcription, allowing RT to switch templates should it encounter a break or damage within the RNA strand being copied. The forced copy-choice model of recombination (5) predicts that the process occurs during minus strand DNA synthesis. However, because template switching also occurs in the absence of breaks in the RNA and may not always be forced, the process fits the more generalized copy-choice model (6). Such interstrand template switching involving two co-packaged nonidentical RNAs would then result in a recombinant provirus. Recombination during plus strand synthesis is described by the strand displacement-assimilation model (7,8) which proposes that two minus strand DNAs are made by one virion and that plus strand synthesis is discontinuous. Although recombination does occur during both minus and plus strand DNA synthesis, studies indicate that minus strand synthesis is the predominant pathway (9 -13).
Recombination occurs at a high rate during retroviral replication (2, 14 -17). The process is predominantly homology driven and can also involve exogenous and endogenous viral RNAs (18 -20). In the case of HIV, reverse transcription-mediated recombination provides an important source of genetic diversity, contributing to the generation of quasispecies within the infected individual (21)(22)(23). It disperses resistance mutations arising during drug therapy (24,25) as well as those randomly introduced by the highly error-prone RT. Furthermore, the process can yield mosaic genomes carrying genetic information from different subtypes (26 -28). Recombination also serves as a rescue mechanism for the virus in situations where RT encounters a damage or defect within the genome. Direct evidence for this is the rescue of replication-defective virus through recombination with endogenous retroviral sequences (18, 20, 29 -31).
In retroviruses, the diploid genome is noncovalently linked near the 5Ј end and is packaged within the virion in this dimeric state (32). After release of the particle from the host cell, it is maintained in a mature and stable dimeric form via interactions with the nucleocapsid protein (33,34). The dimeric state of the retroviral genome is thought to play an important role at several steps in the life cycle, including packaging (35,36), translation (35,37), and recombination (4,20,38,39). The primary linkage structure referred to as the dimerization initiation sequence (DIS) is presented as a hairpin that exposes a palindromic sequence within the loop (40 -43). This structure is located between the primer-binding site (PBS) and splice donor (SD) site within the 5Ј-UTR. Dimerization initiates through Watson-Crick base pairing between the autocomplementary loop sequences of two DIS hairpins, an interaction referred to as kissing (40,41,(43)(44)(45). Subgenomic RNAs containing this region of the HIV-I genome dimerize spontaneously and efficiently in vitro in the absence of viral and cellular proteins (37,46,47). Disruption of the palindromic loop sequence or deletions of the DIS stem loop obstruct dimerization (41,48). The viral nucleocapsid protein, although dispensable for RNA dimerization, can enhance the process (46,49). The DIS hairpin is dispensable for virus replication, although deletion of the element is associated with reduced encapsidation, minus strand synthesis, and infectivity, as well as altered replication kinetics (50,51). In addition to the DIS hairpin, the HIV-I 5Ј leader sequence encodes multiple cis-acting elements, including the packaging signal () and splice donor site, that regulate distinct steps in the virus life cycle (52) (Fig. 1). Whereas the dimeric state of the genome allows for recombination events during reverse transcription, the specific role of template-template interaction sites and genome dimerization in reverse transcription and recombination begs a more detailed analysis.
Studies presented by the Pedersen group have shown recombination-mediated rescue of PBS-defective murine leukemia virus through template switching within the dimer linkage structure (20,(53)(54)(55). These results prompted us to examine strand transfers within the highly structured HIV-I 5Ј leader sequence and the effect of DIS in promoting the process. We hypothesized that direct template-template interactions serve as an important determinant in promoting transfers by bringing homologous regions of the RNA into close proximity. Using RNA templates comprising the dimerizing HIV-I leader sequence and a control nondimerizing pol sequence, we have addressed the role of kissing interactions in facilitating transfers by examining template dimerization, transfer efficiencies, and crossover distribution. The UTR templates supported a high efficiency of crossovers, with a clustering of transfers around the dimerization site. Inhibition of donor-acceptor interactions through deletion of the DIS sequence resulted not only in a suppression of the hot spot but, more significantly, in an overall reduction in transfers across the whole template. Parallel experiments with the pol templates, in which the presence of the DIS sequence induced template dimerization and enhanced transfer efficiency, further verified these observations. Results give insights into the role of template dimerization as a determinant in promoting template switching during minus strand synthesis by HIV-I RT. Plasmid Constructs-Genomic sequences from the NL4-3 and JRCSF strains of HIV-I were amplified by PCR and cloned into pBluescript II KS(ϩ) (Stratagene) to create donor and acceptor constructs for generation of RNA templates. The following PCR primer sets were used in the construction of the various constructs: (a) 5Ј-UTR donor construct pNL187-384, primers SacI/PBS-187 (5Ј-CTGGAGCTCCCGAACAGG-GAC TTGAAAGCG) and BamHI/384 (5Ј-GTGGGATCCCCCATTTATC-TAATTCTCCCC), (b) 5Ј-UTR acceptor construct pJRC150-363, primers SacI/PBS-150 (5Ј-CTGGAGCTCCCTTTTGTCAGTGTGGAAAATCTC) and BamHI/363 (5Ј-GTGGGATCCCGCTTAATACCGACGCTGTCG), (c) pol donor construct pNL-RT3612-3773, primers SacI/RT-3612 (5Ј-G-ACGGAGCTCGCAAGAATGAAGGGTGC) and BamHI/RT-3773 (5Ј-G-CCAGGATCCGCTTGCCAATACTCTGTCC), and (d) pol acceptor construct pJRC-RT3541-3753, primers SacI/RT-3541 (5Ј-GACGAGCT-CCAGGGGCAAGGCCAATG) and BamHI/RT-3753 (5Ј-GCACGGATC-CCCATGCTTCCCATGTTTCC). Restriction enzyme sites within the primers are underlined. PCR fragments were digested with SacI and BamHI and cloned into pBluescript II KS(ϩ) (Stratagene).

Reagents-Recombinant
The DIS mutant JRCSF acceptor, p⌬DIS-NL43 donor, DIS-pol, and DIS-2-pol constructs were generated using the overlap PCR approach. For the DIS mutant JRCSF acceptor, primers SacI/PBS-150 and JRCSF-5Ј-UTR/DISϪ (5Ј-CCACGCCTCTTGCTGTGCGCGCTTCAGC-AAGCCGTGTCC) were used to generate the 5Ј 200-base pair fragment, whereas primers BamHI/363 and JRCSF-5ЈUTR/DISϩ (5Ј-GGACACG-GCTTGCTGAAGCGCGCACAGCAAGAGGCGTGG) were used to amplify the 100-base pair fragment from the 3Ј end. Sequences in bold indicate substitutions, whereas underlined sequences represent regions of overlap between the two fragments. Amplified fragments were purified and used as templates in a second round of PCR with primers SacI/PBS-150 and BamHI/363. A similar approach was used for the rest of the constructs, using the following primer sets: RNA Templates-RNA templates were generated by in vitro run-off transcription from BamHI linearized plasmids using T7 RNA polymerase as per the manufacturer's protocol. Internally labeled transcripts were generated by including [␣-32 P]CTP (3000 Ci/mmol) (NEN Life Technologies) in the transcription mix. All full-length transcripts were purified on a 6% denaturing polyacrylamide gel. All RNA preparations were labeled with 32 P and analyzed by denaturing polyacrylamide gel electrophoresis to ensure intactness of the templates.
Reverse Transcriptase Assays-All RT assays were performed in a 12-l final reaction volume. For transfer assays, 2 units (50 ng) of HIV-I RT were preincubated for 2 min at room temperature with 2 nM template termini of primer-template and 8 nM acceptor in 50 mM Tris-HCl, pH 8.0, 80 mM KCl, and 1 mM dithiothreitol. Acceptor RNA was excluded in donor extension assays. Reactions were initiated by the addition of MgCl 2 and deoxynucleoside triphosphates at a final concentration of 6 mM and 50 M, respectively. After incubation at 37°C, reactions were terminated at the indicated times by the addition of 12 l of termination buffer (90% formamide, 10 mM EDTA, pH 8.0, and 0.1% each of xylene cyanole and bromphenol blue). Reaction products were resolved on 6% polyacrylamide-urea gels and visualized by phosphorimaging using ImageQuant software (Molecular Dynamics).
Analysis of Transfer Products-Transfer assays were performed for 1 h, and then transfer products were purified by polyacrylamide gel electrophoresis and isolated as described previously (56). Alternately, transfer reactions in triplicate were treated with 2 units of E. coli RNase H (Life Technologies, Inc.) to remove residual RNA, phenol chloroform-extracted, and ethanol-precipitated. Transfer products were amplified by PCR using primers SacI/150 (5Ј acceptor primer) and BamHI/384 (3Ј donor primer) for the 5Ј-UTR products and primers SacI/RT-3541 (5Ј acceptor primer) and BamHI/RT-3773 (3Ј donor primer) for the pol products. Purified fragments were cloned into pBluescript II KS(ϩ) and plated to yield clones representing the transfer recombinants. Individual clones were sequenced using M13 (Ϫ20) primers by automated sequencing to identify crossover sites within the recombinants. Both approaches yielded similar results upon sequence analysis.
RNA Dimerization Assays-15 ng of internally labeled RNAs in RT reaction buffer were incubated at 80°C for 1 min and slow-cooled to room temperature. Refolded RNA samples were mixed with 2 l of 50% glycerol, resolved on 4% nondenaturing polyacrylamide gels, and visualized and analyzed by phosphorimaging using ImageQuant software.

RESULTS
Because dimerization interactions within the 5Ј leader sequence bring homologous regions of the two RNAs into close proximity, we hypothesized that this would create a favorable site for template switching by RT during minus strand synthesis. We therefore examined reverse transcription through the HIV-I 5Ј leader sequence using donor and acceptor RNAs comprising the genomic sequence from the PBS to the start of the gag coding region.
Template Switching Is Supported by the 5Ј Leader Sequence-The HIV-I 5Ј-UTR sequence chosen here to examine strand transfers spans the region from the PBS to the start of the gag coding region. This structure simulates the natural replication intermediate during completion of minus strand DNA synthesis. The donor and acceptor templates share a 177-nt-long region of homology (Fig. 1). Synthesis was initiated on the donor RNA using the 5Ј end-labeled MB20 DNA primer that annealed to the 3Ј end of the donor, but not to the acceptor. Transfers result when the extending primer terminus, following release from the donor template, anneals to the homologous acceptor template and completes synthesis.
To test for recombination events, primer extension on the donor was compared in the absence ( Fig. 2A) and presence (Fig.  2B) of acceptor template. Full-length donor-directed synthesis yielded a 214-nt product, which was detected as early as 1 min ( Fig. 2A). In the presence of acceptor RNA, an additional band corresponding to the expected 250-nt transfer product was also visible (Fig. 2B). At 60 min, 18 -20% of full-length products were the result of synthesis completed on the acceptor after transfer, indicating efficient strand transfer synthesis on the UTR templates. In contrast to the rapid generation of donor extension products, a substantial delay was evident in the appearance of transfer products, which were detectable only by 5 min (Fig. 2B, lane 1). The stable secondary structures and sequence characteristics of this template (57, 58) most likely induce the observed pause sites during synthesis.
We note here that the 250-nt product would consist of transfer products generated from single and odd numbers of crossovers, where synthesis was completed on the acceptor. Sequence analysis of transfer products revealed that the majority of products resulted from single crossovers. An average of 1 in 20 recombinants was the product of triple crossovers. Recombinants resulting from double and even numbers of transfers would complete synthesis on the donor template and would therefore be indistinguishable from donor extension products. Sequence analysis of donor extension products, however, did not detect even number transfer products. Apparently, such products, if generated, are undetectable among the large excess of products of uninterrupted synthesis on the donor template.
Template Switching Occurs Predominantly within the DIS Hairpin Region-Because frequent template switching was evident within the leader sequence, we could then correlate the distribution of transfer sites with structural features of the sequence. Cell culture studies in a murine leukemia virusbased system have shown recombination-mediated rescue of PBS-defective virions in which crossovers map to hairpins within the dimerization domain. If transfers at this site were Regions and nucleotide positions in the genomic RNA that correspond to the donor and acceptor RNA templates are indicated. Bold lines represent the NL4-3 donor (black) and JRCSF acceptor (gray) RNA templates. Hatched lines at the template ends denote plasmid-derived sequences on the templates. Open circles on the templates correspond to relative positions of natural, single-nucleotide base differences between the two templates and are designated as markers 1-6. Markers 5 and 6 each comprise two nucleotide variations. Synthesis was primed from the 20-nucleotide DNA primer MB20 (black arrow) annealed to the 3Ј end of the donor.

FIG. 2. Minus strand synthesis catalyzed by HIV-I RT on the 5-UTR leader sequence.
Synthesis was initiated on the donor template using a 5Ј end-labeled MB20 DNA primer in the absence (A) and presence (B) of acceptor RNA. Reactions were terminated at various time points as indicated, and products were resolved on a 6% polyacrylamide gel under denaturing conditions. Full-length synthesis on the donor resulted in a 214-nt donor extension product (A and B), whereas products resulting from template switching and synthesis on the acceptor generated a 250-nt transfer product (B). Pause sites corresponding to the SD and DIS elements are indicated. Lane M, 10-base DNA ladder.
promoted by a specific mechanism involving RNA sequence or structure within this region, we could anticipate a preferred site of template switching or a recombination hot spot within the HIV-I leader sequence. To determine this, the transfer products were cloned and analyzed by sequencing. Donor and acceptor RNAs were generated from HIV-I strains NL4-3 and JRCSF, respectively, allowing the use of natural sequence variations within the homologous region as markers in mapping crossovers. This approach circumvents the need to introduce base substitution markers that might disrupt natural structural features in the RNA that contribute to the pattern of recombination.
The transfer distribution, as analyzed from a total of 90 clones from two independent experiments, is presented in Fig.  3A. As discussed previously, the majority of transfer products (ϳ95%) resulted from single template switching events. Only single crossover products were used in the analysis, although the inclusion of multiple crossover products did not change transfer distribution. The six sequence markers distributed across the region of homology were used in determining crossover sites within the transfer products. Transfers occurred across the length of the template between all of the markers, suggesting that almost any region within the template had the potential to support homologous recombination. A highly preferred region of template switching, however, was clearly evident between markers 3 and 4. A striking 53% of transfers occurred within this 48-nt region. In contrast, adjoining sequences within the template supported a significantly lower but almost uniform frequency of transfers. The three template segments, namely, the 58-nt sequence downstream to marker 1, the 40-nt sequence between markers 1 and 3, and the 30-nt sequence past marker 4, each supported an average transfer frequency of 15-17%. This corresponds to a third of the transfer frequency observed in the segment between markers 3 and 4. To account for the uneven distances between the markers, transfer frequencies within marker segments were corrected by adjusting for the distance between the markers (Fig. 3B). Transfer frequency within a given marker segment was divided by the length of that segment, giving a normalized value that relates to transfer frequency in percentage/10 nt. It is evident from the raw and adjusted distributions (Fig. 3, A and B) that the higher transfer frequency is not a simple consequence of increased marker length. The segment between markers 3 and 4 had three times the potential to promote transfers relative to its neighboring regions.
Interestingly, the preferred region for transfers, defined by markers 3 and 4, includes a major part of the DIS hairpin, the structural element involved in genome dimerization. The template also harbors several other well-characterized structural elements, including the SD hairpin between markers 2 and 3 and the ⌿ and AUG hairpins downstream of marker 1 (Fig. 1). The presence of these structural elements in the donor and acceptor RNAs was confirmed using an RNA folding algorithm (59). Whereas these regions contributed to transfers, they did not create a significant peak of transfers, unlike the DIS. Also, the loop-loop kissing interaction characteristic of the DIS hairpin is not promoted by any of the other hairpin elements within the leader sequence. The biased transfer distribution within the UTR suggests a hierarchy within secondary structures in facilitating transfers. Structural features of the DIS hairpin, including those related to dimerization, appear to be more effective than hairpins in adjacent sequences at promoting recombination.
To test for PCR-mediated recombination, separate extension reactions were performed using NL4-3 and JRCSF acceptor RNAs. Samples from the two extension reactions were mixed, amplified by PCR, and cloned. Sequence analysis revealed that ϳ1 in 24 amplification products was recombinant (ϳ4%), and these did not distribute with a noticeable trend. Recall that all experimental amplification products derived from the transfer product band were recombinants. This control experiment indicates that artifacts from PCR amplification were not significant enough to skew analysis of the transfer products.
DIS Mutant Acceptor-The 98-nt sequence downstream of marker 3, comprising the first half of the template and carrying the AUG, , and SD hairpins, contributed to only 33% of the total internal transfers on the template (Fig. 3). In comparison, the 88-nt region between markers 1 and 4 carrying the DIS and SD stem loops contributed to about 70% of the total internal transfers on the template, three-fourths of which occurred within a short 48-nt sequence harboring the kissing hairpin. The template segment harboring the DIS sequence thus appears to have a pronounced effect in promoting transfers. We hoped that a finer analysis of transfer distributions at this preferred region would provide insights into the mechanism of the transfer process. To address this, the JRCSF DIS mutant acceptor carrying an additional set of markers 3a and 3b at the base of the DIS stem loop was generated (Fig. 4A). To minimize possible structural distortions, this was done by substituting the U-A base pair at the base of the stem with A-U, so that the stability of the DIS hairpin was minimally compromised. Use of the RNA folding algorithm (59) revealed similar folding potentials for both wild type (WT) and DIS mutant acceptor templates. Transfer assays were performed using the NL4-3 donor and JRCSF DIS mutant acceptor, and transfer efficiency and distribution were analyzed as described previously.
Transfer efficiency measured on the mutant acceptor was similar to that measured on the WT acceptor. Fig. 4B shows the distribution of crossovers supported by the mutant acceptor. Data were averaged from two independent experiments, each consisting of 65-70 single crossover products. The new markers (3a and 3b) effectively flank the DIS hairpin and distinguish transfers within and outside of this structure (Fig. 4A). Several interesting conclusions are evident. Firstly, the overall distribution of transfers on the mutant acceptor showed a trend almost identical to that on the WT acceptor (compare Figs. 3 and 4B). The preferred site between markers 3 and 4 supported 49% of the transfers, whereas transfers at other regions were relatively uniform and low. Compared with transfers on the WT acceptor, a slight decrease in transfer frequency was observed between markers 3 and 4 on the mutant acceptor, and a slight increase was observed between markers 2 and 3 and within the 31-nt sequence 5Ј of marker 4. This could be due to structural changes induced by the base substitutions in the mutant acceptor. However, the variation was minimal and did not change the overall trend.
Secondly, in examining the original hot spot between markers 3 and 4, we noted that of the 49% of transfers that mapped within this region, two-thirds occurred just past the kissing hairpin (between 3b and 4), whereas only one-third occurred within the hairpin itself (between markers 3 and 3b). Similarly, of the 20% transfers between markers 2 and 3, only 5% mapped within the DIS 3Ј stem sequence, and the remaining 15% occurred outside of the DIS. Contrary to our expectations, the preferred site did not map within the DIS hairpin (between 3a and 3b). Instead, crossovers were predominantly detected within the region immediately 5Ј of the kissing stem loop. The data clearly suggest that transfer of the primer terminus occurred after synthesis through the kissing hairpin. Two simple transfer mechanisms can be envisioned in this scenario: one in which the transfer process is initiated at the primer terminus, and a second in which the transfer process is initiated via primer-acceptor interactions behind RT and the primer terminus.
Deletion of the DIS Element Impedes Dimerization and Lowers Transfer Efficiency-If the dimerization signal were directly responsible for creating the observed preferred site for template switching, then deletion of this signal should have a noticeable effect on both dimerization and transfers. To put this argument to the test and therefore address the role of the DIS hairpin in promoting transfers, the ⌬DIS donor template was constructed. In this template, a 23-nt sequence comprising the upper stem loop region of the hairpin was deleted.
Wild type and modified templates were assayed for dimerization efficiency by heating the individual templates and slowcooling them (Fig. 5A). As observed, both wild type donor and acceptor templates dimerized with similar efficiencies between 18% and 20% (Fig. 5A, lanes 1 and 2). The base-exchanged DIS mutant acceptor template was also able to dimerize, albeit with a slightly lower efficiency than the WT templates (Fig. 5A, lane   3). Unfortunately, the close sizes of the donor and acceptor RNAs made it difficult to distinguish donor-acceptor heterodimers from homodimers by this analysis. However, because both templates are almost identical in sequence, their ability to self-dimerize is indicative of their potential to heterodimerize with each other. As anticipated, deletion of the dimerization signal impeded the ability of the ⌬DIS template to self-dimerize (Fig. 5A, lane 4). This is in agreement with previous studies characterizing the role of the DIS hairpin in promoting the dimerization of the UTR region (48).
To address the role of the dimerization signal in promoting transfers, we performed transfer assays using the ⌬DIS donor in the presence and absence of WT acceptor (Fig. 5, B and C). Transfer products of two different sizes were generated, with the size depending on the region where transfers occurred. Primers that transferred before the deletion site on the donor generated the 250-nt product, whereas those that transferred after the deletion site produced the shorter, 227-nt transfer product (Fig. 5, C and D). Whereas the dimerizing template promoted efficient transfers (18 -20%), a 4-fold reduction in transfer efficiency (5-6%) was observed when synthesis was initiated on the nondimerizing ⌬DIS donor (Fig. 5D). In addition to the overall decrease in transfer efficiency, a significant delay in accumulation of transfer products was also noticeable with the ⌬DIS donor as compared with the WT donor (Fig. 5, C  and D). Taken together, these results demonstrate a positive correlation between the DIS-mediated dimerization and recombination potential within the UTR templates.
Deletion of the DIS Element Results in Suppression of the Hot Spot-Deletion of the DIS hairpin reduced both the ability of the template to dimerize and the ability of the template to promote transfers. To test how these effects impacted the transfer distribution, transfer products were amplified from the whole reaction mix, cloned, sequenced, and analyzed as described previously. Fig. 6A summarizes the transfer distribution as analyzed from two independent experiments. In the WT donor, about one-third of the total transfers were concentrated within a short region adjacent to the DIS and therefore contributed to one-third of the 18% transfer efficiency measured on the template (Fig. 4B). On the DIS deletion donor, two template segments (a segment before marker 1 and the ⌬-4 segment) each accounted for about one-third of the total transfers. Considering the 5-6% transfer efficiency promoted on the entire ⌬DIS, each of these segments represents only about 1.5-2% transfer efficiency. In the absence of the DIS, the hot spot was clearly subdued, and a more even distribution of transfers was noticed. As observed from the transfer distribution and crossover frequencies, the reduced transfer efficiency measured on the ⌬DIS template could not be accounted for by the simple loss of the hot spot. The absence of the DIS hairpin appears to have reduced transfers across the whole length of the template in addition to suppressing the original hot spot 5Ј to DIS. Synthesis was initiated on the donor template using the 5Ј end-labeled MB20 DNA primer in the absence (B) and presence (C) of WT acceptor RNA. Reactions were terminated at the indicated time points, and products were resolved on a 6% polyacrylamide gel under denaturing conditions. Full-length synthesis on the donor resulted in a 191-nt donor extension product (B and C), whereas products resulting from template switching and synthesis on the acceptor generated transfer products of two different sizes (C), shown in detail in D. The larger, 250-nt product resulted from transfers occurring before the deletion site on the donor, whereas the 227-nt product resulted from template switching after the deletion site. Lane M, 10-base DNA ladder. D, enlarged view of transfer product accumulation in assays containing the WT UTR donor template (left panel) and the ⌬DIS donor template (right panel). Both assays contained WT acceptor RNA. Reactions were sampled at 5, 10, 15, 30, 45, and 60 min. Donor extension and transfer products are indicated. The two different-sized transfer products (250 and 227 nt) observed in the ⌬DIS donor assays resulted from crossovers occurring before and after the deletion site in the donor template, as indicated by the schematics to the right. When template switching occurred before the deletion site, the DIS region was copied from the acceptor, generating the 250-nt transfer product. Transfers occurring after the deletion site generated the shorter, 227-nt product lacking the DIS sequence. The sum of both products was used in calculating the transfer efficiency. Transfer efficiencies measured on the two templates are indicated at the bottom of each panel.
To adjust for the nonuniform marker distances, transfer frequencies within each marker segment in Fig. 6A were divided by the length of the segment. The corrected transfer frequency in percentage/10 nucleotides would then be 6.1 (59-nt segment before marker 1), 5 (segment 2-⌬), 9.9 (segment ⌬-4), and 4.9 (segment 4). Although the region 5Ј to the deletion site showed a higher frequency of transfers, it was only about 50% higher than the region upstream of marker 1. In contrast, comparing the same segments in the UTR template, the segment 5Ј to the DIS displayed a 3.7-fold higher frequency of transfers. The effect of the DIS deletion became more apparent and more distinct when the transfer efficiency was also taken into account. To account for the 3.5-fold lower transfer efficiency supported by the ⌬DIS UTR as compared with the WT UTR, we divided the distance-adjusted numbers by 3.5 (Fig.  6B). The data representation in Fig. 6B thus allows comparison of recombination frequency between templates containing and lacking the DIS sequences (represented in Figs. 3B and 6B, respectively) after correcting for both marker spacing and transfer efficiency. Whereas in the WT template, transfer frequencies were high around the dimerization site and peaked at the DIS, the ⌬DIS showed an overall low frequency of transfers across the templates, with a redistribution of transfers and suppression of the original preferred site. Thus, a comparison of the transfer efficiencies and regional crossover frequencies among the various WT and mutant UTR sequences strongly suggests that the DIS sequence contributes to creating a preferred transfer site within the HIV-I 5Ј-UTR.
The DIS Hairpin Induces Efficient Dimerization of the Ectopic pol Templates-To test the potency of the DIS hairpin and to allow comparison with the dimerizing UTR templates, we generated a set of control donor and acceptor templates using the pol coding region of the genome. Whereas this region of the genome has no known role in genome dimerization, it would have its own flavor of sequence and secondary structures. The pol donor and acceptor templates share a 140-nt region of homology, with nucleotide base differences distributed at seven positions within this region (Fig. 8A). To test the competence of DIS-induced dimerization in promoting transfers, we presented it within this ectopic pol sequence. The DIS-pol donor and acceptor (Fig. 8C) contain the upper stem loop (28 nt; Fig.  4A) of the kissing hairpin, whereas the DIS-2-pol template set (Fig. 8D) contains the entire kissing hairpin (35 nt).
Wild type and DIS insertion pol templates were assayed for dimerization efficiency by heating the internally labeled templates individually and slow-cooling them (Fig. 7A). As tested under RT assay buffer conditions, the pol donor dimerized very inefficiently (Ͻ1%), whereas ϳ10% of the acceptor existed as dimer (Fig. 7A, lanes 1 and 2). In contrast, both the DIS-pol and DIS-2-pol templates dimerized with very high efficiencies (Fig.  7A, lanes 3-6). In all template sets, the acceptor dimerized with higher efficiency than the donor. Structures or sequences within the extended 5Ј region of the acceptor most likely contribute to this effect. When donor and acceptor RNAs were present in equimolar concentrations, both the DIS-pol and DIS-2-pol templates also formed heterodimers in addition to the homodimers (Fig. 7A, lanes 8 and 9), suggesting that DISinduced dimerization was facilitated within the ectopic pol sequence. In the absence of the kissing hairpin, no detectable levels of heterodimers were observed with the WT pol templates (Fig. 7A, lane 7). Unlike the UTR templates, the significant size differences between the pol donor and acceptor templates made it possible to distinguish donor-acceptor heterodimers from homodimers.

DIS-induced Dimerization Enhances Transfer Efficiency in the pol
Templates-To address the role of the dimerization signal in promoting transfers, we performed transfer assays using the WT and DIS-containing pol template sets under assay conditions similar to those used for the UTR templates (Fig. 7, B-D). Quantitative analysis of transfer products yielded 6 -7% transfer efficiency for the pol templates, whereas the DIS-pol and DIS-2-pol templates showed transfer efficiencies of 11-12% and 10 -11%, respectively, approximately doubling the efficiency. Insertion of the 28-and 35-nt DIS sequences increased homology length of the original template by 16% and 25%, respectively. However, this cannot account for the almost 200% increase in transfer efficiency. Additionally, as observed previously with the UTR templates (Fig. 5C), an increased delay in accumulation of transfer products was also observed with the nondimerizing pol templates as compared with the dimerizing templates (compare Fig. 7B with Fig. 7, C  and D). Quantitative analysis by phosphorimaging detected transfer products by 2.5 min with both dimerizing templates, whereas in the pol templates, transfer products were detectable only by 5 min. With all three substrates, full-length donor extension products were detectable as early as 1 min.
To better understand how the presence of the DIS sequence relates to recombination, we sequenced the transfer products and analyzed transfer distributions in the three sets of templates (Fig. 8). The unaltered raw transfer distributions are presented as averaged from a minimum of two independent experiments (Fig. 8, A, C, and D). Transfer distributions within the various pol templates were normalized for the unequal marker distances to examine the relative transfer potentials of the different segments. To arrive at this, the transfer frequencies (represented in percentages in Fig. 8, A, C, and D) within marker segments were divided by the segment lengths. To allow a vertical comparison between the templates, the distance-corrected frequencies in the pol templates were addition-FIG. 6. Transfer distribution on the ⌬DIS UTR template. A, average transfer frequencies within the template segments as determined from two independent experiments. The data represent the number of clones that indicated a transfer event at a given segment presented as a fraction of the total number of single crossover clones analyzed. B, corrected transfer frequencies after normalizing the raw distribution from A for distance and the 3.5-fold lower transfer efficiency of ⌬DIS compared with WT UTR (Fig. 5C; 18-20% for WT versus 5-6% for ⌬DIS; see "Results" section for details). Numbers in A were divided by the respective segment length to give transfer frequency/10 nt. These were as follows: segment 1, 6.1; 2-D, 5; D-4, 9.9; and after 4, 4.9. These numbers were then divided by 3.5 to arrive at the data presented in B. This allowed comparison of recombination frequency between templates containing and lacking the DIS sequences (represented in Figs. 3B and 6B, respectively). Values were averaged from two independent experiments of about 40 clones each and were based on products resulting from single crossover events. Description and calculations are similar to those discussed in the Fig. 3 legend. ally normalized for the 2-fold lower transfer efficiency promoted by this template compared with the dimerizing DIS and DIS-2-pol templates (Fig. 8, B, E, and F). The effectiveness of the DIS sequence in facilitating transfers, even when presented within an ectopic region, clearly demonstrates a positive correlation between DIS-mediated template dimerization and recombination potential.
It is immediately apparent from the transfer distributions that unlike the case of the UTR template (Fig. 3A), transfers within the unaltered pol template did not highlight any preferred sites (Fig. 8A). About 40% of the extending primers transferred before marker 3. When transfer frequencies were normalized for distance, segment 3-4 stood out as the most recombinogenic segment, followed by segment 2-3 and the end segment after 5 (Fig. 8B). Insertion of either the partial or the full DIS hairpin caused some changes in distribution (Fig. 8,  C-F). Examination of the normalized frequencies showed two concentrations of crossovers, one centered around the DIS and the other at the 5Ј end of the segment (Fig. 8, E and F). In both templates, the end segments promoted the most transfers, followed by the D-3b (DIS loop and 5Ј stem) segment. In the DIS-2-pol template, the 13-nt 3Ј DIS stem does not appear as potent as the shorter, 9-nt stem in the DIS-pol. This is most likely because the effect is diluted out in the longer 3a-D segment (5.6% in 9 nt versus 4.4% in 13 nt). As observed with the WT UTR template, the presence of the DIS (both the partial and full sequences) enhanced transfers across the whole template. The concentration of crossovers near the DIS is also consistent with the characteristics of the UTR template. Enhanced transfers at the pol 5Ј end may be induced by an additional template-template interaction site promoted in the presence of the DIS. Overall, the ability of the DIS sequence to induce template dimerization and the correlation with transfer efficiency and distribution in both UTR and pol templates strongly suggest that DIS-induced dimerization enhances recombination.

DISCUSSION
Extensive analyses both in cell culture and in vitro have identified substrate and enzyme features that are determinants in driving retroviral recombination. To date, several studies have shown pausing as a strong facilitator of transfers during minus strand synthesis (60 -62). Stalling concentrates RT-associated RNase H cuts. Invasion by the acceptor at the gap produced by RNase H cleavage leads to displacement of the donor template and primer capture (60, 63-65). Alternately, the nascent DNA may be released from the donor, which can then interact with the acceptor from solution. Conditions that promote pausing are  1, 3, and 5) and acceptor (lanes 2, 4, and 6) RNAs were individually refolded under standard transfer assay conditions and subjected to native polyacrylamide gel electrophoresis (see "Materials and Methods" for details). The slower migrating bands (upper bands) observed with the pol (lanes 1 and 2), DIS-pol (lanes 3 and 4), and DIS-2-pol (lanes 5 and 6) templates correspond to RNA homodimers, whereas the faster migrating bands correspond to the monomer fractions. Dimerization efficiencies are indicated below each lane and were calculated as the percentage of dimer from the total of dimer and monomer. Equimolar concentrations of donor and acceptor templates were mixed and refolded together (lanes [7][8][9]  therefore conducive to template switching. Misincorporation during synthesis promotes strand transfers primarily through such a pause and RNase H-mediated mechanism (66,67). Similarly, homopolymeric runs within the templates that are hot spots for frameshift mutations (68,69) also affect recombination (70). In addition to pausing, the role of template-template interactions as an alternate mechanism in driving transfers is becoming increasingly evident (53,56). The efficiency of such a mechanism is very convincingly demonstrated at the dimerization site within the retroviral genome (20,(53)(54)(55). Whereas structure and sequence features of the template present favorable sites for template switching, properties of RT are also critical in facilitating the process. The absolute requirement for RT-associated RNase H in facilitating strand transfers on heteropolymeric RNA templates has been demonstrated by a number of independent studies (60, 63-65, 71).
Using the HIV-I TAR, Kim et al. (56) demonstrated that stable hairpin structures promote recombination not only at the pause site at the hairpin base but also within the stem loop. Lessons from the TAR hairpin provided the first clues for the possible role of template-template interactions in driving recombination. To test this more extensively, we chose a region of the HIV-I genome that is well characterized for its ability to dimerize, the DIS containing 5Ј leader RNA. Dimerization via DIS-mediated kissing interactions has been demonstrated to be very efficient, stable, and promoted in the absence of any protein components.
We initiated experiments with the UTR templates with the following hypothesis: if template interactions facilitate transfers, then the dimerization site should be a preferred region for recombination. In fact, not only did the UTR templates support a high efficiency of template switching during reverse tran- FIG. 8. Transfer distribution on the pol and DIS-containing pol templates. Details of the schematic are the same as those described in the Fig. 1 legend. Nucleotide positions in the genomic RNA that correspond to the donor and acceptor pol templates are indicated (A). Synthesis was primed from the 19-nucleotide DNA primer MB22 (black arrow) annealed to the 3Ј end of the donor. Numbers 1-7 designate single-nucleotide base differences between the donor and acceptor templates. Marker 3 comprised a 2-base variation. A, C, and D, transfer distribution showing the average transfer frequencies within marker segments in the pol (A), DIS-pol (C), and DIS-2-pol (D) templates as determined from two independent experiments. Data represent the number of clones that indicated a transfer event at a given segment presented as a fraction of the total number of single crossover clones analyzed. The 28-and 35-nt DIS sequence insertions within marker 3 are indicated for the DIS-pol (C and E) and DIS-2-pol (D and F) templates, respectively. Marker D represents the single-nucleotide base variation within the NL4-3 and JRCSF DIS sequence. B, D, and E, corrected transfer frequencies after normalizing the raw distribution from A, C, and D, respectively, for marker distance and the 2-fold lower transfer efficiency of the pol template compared with DIS-containing templates (6 -7% for WT versus 11-12% for DIS-pol and 10 -11% for DIS-2-pol; see "Results" for details). Corrected frequencies for DIS-pol (E) and DIS-2-pol (F) templates were derived by simply dividing the frequencies within the marker segments from C and D by the respective segment length to give transfer frequency/10 nt. Distance-corrected frequencies similarly derived for the pol templates were as follows: segment 1, 1.2; 1-2, 3.4; 2-3, 11; 3-4, 13.6; 4 -5, 5.7; and after 5, 9.7. To account for the almost 2-fold lower transfer efficiency of the pol template, these distance-corrected numbers were further divided by 2 to arrive at the data in B. This allowed comparison of recombination frequency between the templates containing and lacking the DIS sequences. Values were averaged from two to three independent experiments of about 40 clones each and based on products resulting from single crossover events. Description and calculations are similar to those discussed in the Fig. 3 legend.
scription, but 53% of the transfers were concentrated within a 48-nt segment encompassing the dimerization signal, consistent with a template interaction-mediated transfer mechanism.
The effect of the DIS hairpin became more evident when transfer efficiency and distribution were examined in the DIS deletion template (⌬DIS). In the absence of the kissing hairpin, the UTR sequence was not only ineffective in dimerization but also displayed a striking 3-4-fold reduction in transfer efficiency when compared with the WT template. 1 of every 5 of the extending primers transferred on the WT UTR template, whereas only 1 in 18 -20 of the extending primers transferred on the nondimerizing ⌬DIS template. The 48-nt preferred site in the WT template alone accounted for a transfer efficiency of about 10% (53% of 18 -20%, Fig. 3A), which was almost twice the transfer efficiency of the ⌬DIS template as a whole. The segment 3Ј to marker 4 accounted for transfer efficiencies of 1.7% and 6.3% in the absence and presence of the DIS sequence (31.7% of 5.5%, Fig. 6A; compare with 33% of 19%, Fig. 4B). The combined data from Figs. 3, 4, and 6 clearly demonstrate that the preferred transfer site is 5Ј to the DIS and created only in the presence of the DIS sequence. Secondly, as observed from the corrected frequencies, transfers were more evenly distributed on the ⌬DIS template, which did not show any distinct hot spots. Frequencies, in decreasing order, in the ⌬DIS template were 2.8, 1.7, 1.4, and 1.4, whereas in the WT template, they were 11, 4.6, 4.6, and 3. In the WT UTR, segment 3-4 was 2.4 times more recombinogenic than the second most favorable site (segments 2-3 and 4-end) and 3.6 times more recombinogenic than segment 1. However, in the ⌬DIS template, segment ⌬-4 was only 1.6 times more recombinogenic than the second most favorable site, i.e. the segment before marker 1. In the absence of the DIS sequence, there is a redistribution of transfers with a suppression of the original hot spot.
Clearly, the 4-fold drop in transfer efficiency on the ⌬DIS cannot be explained by the simple reduction of the hot spot. There is a decreased frequency of crossovers across the whole template, thereby making the whole region less conducive to recombination. This is clear from the transfer frequencies normalized for distance and decreased transfer efficiency (Fig. 6B). Whereas the distance-corrected frequencies allow comparison of transfer potentials of segments within a given template, the additional correction for transfer efficiency allows comparisons between templates differing in transfer efficiencies. Whereas this does not alter the profile within the given template, it correctly represents the lower transfer potential within the template.
Our conclusions on the role of the DIS sequence in mediating transfers in the UTR templates are further substantiated by the transfer data from the pol templates. The DIS sequence not only induced dimerization of the pol templates but also caused an approximate doubling of transfer efficiency. As with the ⌬DIS UTR templates, the change in transfer efficiency cannot be attributed to the simple change in template length. Clearly, it is the characteristics imparted by the DIS sequence to the template, rather than its length, that is the important determinant. In both sets of templates, dimerization makes the overall region more recombinogenic. Whatever factors contribute to crossovers within the individual segments of the template, be they sequence, pausing, or structure-induced, DISinduced dimerization helps to accentuate them.
The requirement of RNase H for transfers has been demonstrated in a number of previous studies (60,(63)(64)(65)71). In our system as well, the HIV-IRT E478Q mutant (72) lacking RNase H activity was ineffective in producing detectable levels of transfer products with the UTR templates (data not shown). Also, although the UTR template harbors several other struc-tural elements, including the SD, ⌿, and AUG hairpins, these did not appear to be major promoters of transfers. Because the template RNA is continually unwound, cleaved, and subjected to dynamic changes in conformation as synthesis proceeds (73), transient conformations also most likely contribute to the transfer process. Such factors, although difficult to ascertain, cannot be ignored.
Pausing is a strong facilitator of transfers (60 -62). A previous study (57) examining RT pause sites within the HIV-I leader sequence has shown the strongest pausing within the A-U stretch 3Ј to the SD hairpin, with weaker pause sites at other regions. In the present study, weak pause sites were induced both within the poly(A) stretch and start of the SD hairpin and within the DIS hairpin (Fig. 2). Quantitative analysis of these pause sites showed that synthesis was not significantly impeded within these regions, even at the dimer interaction site. This suggests that although pausing is very likely an important factor, it is not the sole facilitator of the transfer mechanism within this region. Clearly, transfers are promoted through the interaction of several factors. This also agrees with the observations of Kim et al. (56), who found that transfers within the TAR hairpin did not correlate with a strong pause site, unlike the pause-induced transfers at the base of the hairpin.
Two interesting characteristics were observed with the transfers promoted in the UTR. Firstly, contrary to our expectation, the preferred site mapped to the region immediately 5Ј of the DIS hairpin. Significantly, distribution data primarily indicate the region within which synthesis is reinitiated on the acceptor after primer terminus transfer. They do not reveal exact details of the transfer process itself. Secondly, a noticeable delay was observed in the appearance and accumulation of transfer products as compared with donor extension products, which was further augmented in the nondimerizing ⌬DIS template. Observed in previous analyses of strand transfer in vitro (64), this lag in appearance of transfer products suggests that one or more steps in the transfer mechanism limit the rate of generation of transfer products. These steps could include formation of a transfer intermediate, the primer transfer process itself, or initiation of synthesis from the transferred primer terminus. Such mechanistic details are currently poorly understood and require more detailed analysis to address and dissect the steps in the transfer process.
Based on the observed crossover distribution within the UTR sequence, together with our current understanding of transfer mechanisms, we propose the following model for RT-mediated template switching at the DIS. As synthesis proceeds, RTassociated RNase H cleaves the donor RNA. The clustering of pause sites within the DIS sequence is indicative of increased RNase H activity within this region. This creates gaps within the hybrid behind the extending primer terminus, where the acceptor-nascent DNA interactions are initiated. Once a significant length of the DNA behind the RT has base-paired with the acceptor, the primer terminus, which has now extended past the DIS, is also forced to transfer, completing the process. Very likely, the primer terminus is displaced from the donor via progressive invasion by the acceptor (74,75). DeStefano et al. (74) have shown the nascent DNA to form initial interactions with the acceptor template over several bases, forming a stable duplex before extension on the acceptor. Such interactions can be favorably promoted within the DIS region. Dimerizationinduced proximity between primer donor and acceptor thus increases the efficiency and frequency of transfers around the dimerizing region relative to other regions.
A recent study in cell culture showed that an average of three crossovers occur per replication cycle in HIV-I (12). The same study also suggested that crossovers occur throughout the length of the genome and with similar efficiency, and no obvious hot spots were seen when large genome segments were compared. The 5Ј-UTR sequence was not analyzed in that study. In our system, the DIS region appears as a hot spot in the context of a short, 200-nt length of the genome but is not likely to be a uniquely prominent hot spot in the context of the 10-kb genome. Recombination events within UTR could serve to correct for defects in regulatory elements such as the PBS and thereby contribute to enhancing viability and fitness.
The focus of the current study was to ascertain the role of template interactions in facilitating transfers and thereby add to our current understanding of factors and determinants that promote recombination in HIV. Based on the observations in this study, we propose that template interactions offer an alternate mechanism for promoting recombination, which is distinct from pause-mediated transfers. Both mechanisms may contribute to creating the most active sites for crossovers. Whereas significant progress has been made in identifying important determinants for recombination, the actual mechanisms of template switching are only just beginning to be understood. Gaining a better insight into the "how and why" of template and enzyme features that impede or promote template switching will allow us to better appreciate recombination as a driving force in retroviral evolution and genetic variation.