Recombination in the 5′ Leader of Murine Leukemia Virus Is Accurate and Influenced by Sequence Identity with a Strong Bias toward the Kissing-Loop Dimerization Region

Abstract

Retroviral recombination occurs frequently during reverse transcription of the dimeric RNA genome. By a forced recombination approach based on the transduction of Akv murine leukemia virus vectors harboring a primer binding site knockout mutation and the entire 5′ untranslated region, we studied recombination between two closely related naturally occurring retroviral sequences. On the basis of 24 independent template switching events within a 481-nucleotide target sequence containing multiple sequence identity windows, we found that shifting from vector RNA to an endogenous retroviral RNA template during minus-strand DNA synthesis occurred within defined areas of the genome and did not lead to misincorporations at the crossover site. The nonrandom distribution of recombination sites did not reflect a bias for specific sites due to selection at the level of marker gene expression. We address whether template switching is affected by the length of sequence identity, by palindromic sequences, and/or by putative stem-loop structures. Sixteen of 24 sites of recombination colocalized with the kissing-loop dimerization region, and we propose that RNA-RNA interactions between palindromic sequences facilitate template switching. We discuss the putative role of the dimerization domain in the overall structure of the reverse-transcribed RNA dimer and note that related mechanisms of template switching may be found in remote RNA viruses.

The dimeric retrovirus RNA genome consists of two plus-strand RNAs which are linked by RNA interactions primarily within the highly structured 5′ untranslated region (5′ UTR). During reverse transcription, genetic recombination occurs frequently (23, 24, 59), thus contributing to viral variability (27) and escape from lethal mutations (60). Recombination may occur during reverse transcription of a heterodimeric genome containing viral RNA of both exogenous and endogenous origin (18). Indeed, various endogenous viral relics in the mouse genome have proved to represent a source of functional sequences that may participate in recombinational mutation repair (12, 13, 36, 38, 39, 45, 56).

Reverse transcriptase-mediated recombination has been demonstrated mainly during RNA-directed minus-strand DNA synthesis, in accordance with the proposed models for forced copy choice (10) and minus-strand exchange recombination (11), but may also involve plus-strand DNA assimilation (26) or unconventional template shifting during plus-strand synthesis (39, 53). Mechanisms for template switching during minus-strand DNA synthesis have been extensively studied by in vitro approaches, and the findings show that reverse transcriptase frequently pauses during polymerization of the nascent minus strand (6, 16, 29, 66). Pausing of the enzyme is directed by sequences or secondary structures within the RNA template (17, 29) and may result in enhanced strand transfer at specific pause sites (6, 16, 66). It was therefore expected in a recent in vitro work that pausing at the human immunodeficiency virus type 1 (HIV-1) transactivation response region causes strand transfer; however, the preferred transfer site was mapped within the stem structure and did not coincide with the pause site (28). From in vivo studies of recombination between marker gene cassettes of nonviral origin, it appears that the length of template sequence identity is a primary determinant in template switching of the nascent DNA strand (68).

The primary dimerization domain, designated the dimer linkage site, has been mapped to the 5′ UTR in a region overlapping the retroviral packaging signal. Support for a kissing-loop-loop interaction at least during initiation of RNA dimerization has recently been provided through in vitro studies of murine leukemia virus (MLV) (20, 52), avian sarcoma-leukosis virus (19), and HIV (9, 14, 30, 35, 44, 48–50, 57). This kissing-loop interaction model proposes base pairing between palindromic loop sequences of the dimer linkage site and subsequent isomerization of the stem-loops (20), thereby generating an interstrand RNA duplex which represents a local antiparallel linkage of the two RNA subunits. However, in vivo investigations suggest that the interaction of palindromic loop sequences is not absolutely required for retroviral replication (4, 8, 22, 31, 47); therefore, alternative yet unknown RNA interactions within or outside the 5′ UTR may support dimerization (4, 31). Recent work suggests that the HIV-1 kissing-loop dimerization region may be essential also for optimal proviral DNA synthesis (47).

By use of a single-cycle vector transfer protocol, we have previously developed a forced recombination system based on the strongly restricted replication of Akv MLV-derived vectors harboring a mutated primer binding site (PBS) sequence and the 5′ 244 nucleotides of the wild-type 476-nucleotide Akv 5′ UTR. PBS-modified vectors may thus be transferred through reverse transcriptase-mediated recombination with a MLV-like endogenous virus (MLEV) involving either R-U5-mediated second-strand transfer (39) or minus-strand template shift within the 5′ UTR (38). In the latter case, we registered a clustering of recombination sites within a narrow region of the 244-bp 5′ UTR coinciding with the primary dimerization site, raising the possibility of a combined role of template sequence identity and RNA secondary structure in template switching between naturally occurring retroviral sequences.

To assess the pattern and precision of retroviral recombination between highly structured natural viral RNAs, we have by forced recombination studied transduction of vectors carrying the full-length 476-nucleotide Akv 5′ UTR. We address in this report (i) whether the overall structure of the entire 5′ UTR and more specifically cis-acting elements in the downstream part of the 5′ UTR influence the pattern of template switching during minus-strand DNA synthesis, (ii) if homologous recombination is a simple matter of donor and acceptor template sequence homology at the transfer site, and (iii) if reverse transcriptase template shifting is governed by a minimum length of sequence similarity between the nascent minus-strand DNA and the RNA acceptor template. We conclude from our studies that the kissing-loop dimerization domain within a large recombination target sequence consisting of multiple sequence identity windows (SIWs) is a hot spot for recombination. On this basis, we propose that close RNA-RNA interactions in the primary dimerization palindrome facilitate template switching.

MATERIALS AND METHODS

Vector construction.

Vectors pPBSPro244ΨAkv-neo and pPBSUmu244ΨAkv-neo have been previously described (33, 38). Vector designations refer to the type of PBS present and 5′ UTR length. Vectors harboring the complete 476-bp Akv 5′ UTR upstream from the Tn5 transposon fragment which encompasses the neomycin phosphotransferase gene (neo) were generated as follows. A fragment containing the 5′ long terminal repeat, (LTR), the proline PBS (PBS-Pro), and the 476-bp Akv 5′ UTR was PCR amplified from pAKR59 (32), which contains wild-type Akv cis-acting elements and coding regions. The amplified sequence was cloned by standard procedures into the appropriate position of pPBS-Pro244ΨAkv-neo; the resulting retroviral vector was designated pPBSPro476ΨAkv-neo. PBS-Pro was replaced by the PBS-Umu modification by a two-step PCR approach previously described (33) to generate pPBSUmu476ΨAkv-neo. Vector constructs are illustrated in Fig. 2A.

FIG. 2.

Vector structure and function. (A) Vectors, expression in packaging cells, and transductional titer. A modified PBS sequence, designated PBS-Umu, was introduced in Akv MLV-derived vectors harboring the neo gene embedded in the bacterial Tn5 transposon. pPBSPro244ΨAkv-neo and pPBSUmu244ΨAkv-neo were previously used in studies of retroviral recombination (38). The entire Akv 5′ UTR was inserted in vectors pPBSPro476ΨAkv-neo and pPBSUmu476ΨAkv-neo. Vectors studied in this work and previously differ in the length of the 5′ UTR sequence, being 244 and 476 bp, respectively. neo expression was estimated by transfection of Ψ-2 packaging cells followed by counting of G418-resistant colonies; the resulting estimate of vector expression is given as 10² G418-resistant colonies per transfection of 10 μg of vector DNA. Transductional titers were measured by counting the G418-resistant CFU per milliliter of medium transferred from stably transfected packaging cells; titers have been normalized to 10⁷ producer cells and represent average values for three independent experiments. ND, not determined. The schematic representation of the secondary structure of the MLV 5′ UTR is based on studies by Tounekti et al. (62) and Mougel et al. (42). Ten putative stem-loops (SLs 1 to 10) are found within the region. As indicated by dotted lines, SLs 1 to 9 and a putative shortened SL 10 were included in the longer vectors harboring the complete 5′ UTR, whereas only SLs 1 to 4 were included in the shorter vectors (38). By analogy with the previously presented model, stem-loops are located as follows (position numbers as given in Fig. 1): SL 1, 236 to 265; SL 2, 291 to 322; SL 3, 329 to 370; SL 4, 372 to 394; SL 5, 400 to 417; SL 6, 423 to 453; SL 7, 459 to 480; SL 8, 486 to 534; SL 9, 539 to 561; and SL 10, 605 to 647. (B) Experimental approach for studies of forced recombination. PBS-mutated vectors were tested in single-cycle transfer protocol by transfection into Ψ-2 packaging cells and subsequent virus transfer to NIH 3T3 target cells. Recombination-based vector transduction, or forced recombination, is the result of (i) heterodimeric RNA encapsidation, (ii) initiation of minus-strand synthesis, (iii) successful plus-strand transfer, and (iv) expression of the neo gene. G418-resistant colonies were individually cloned or pooled (see Materials and Methods) in order to sequence individual transduced proviruses or to allow PCR screening for recombinants, respectively.

Sequence analysis and cloning of the MLEV 5′ UTR.

Sequencing of the upstream part of the MLEV 5′ UTR was performed by various PCR-based strategies as previously described (38). Among them, a semirandom-primed PCR approach (58) was used to selectively amplify glutamine PBS (PBS-Gln)-containing sequences in viral cDNA prepared from total RNA in virus-containing medium from nontransfected Ψ-2 packaging cells. To obtain the sequence of the 3′ part of the 5′ UTR, PCR products were sequenced with a primer matching the degenerate primer linker used in the semirandom PCR. The obtained MLEV sequence allowed the design of primers to specifically amplify the MLEV 5′ UTR. The 465-bp MLEV 5′ UTR was introduced into vector context by PCR connection with a PCR fragment harboring the 5′ LTR and PBS of choice and subsequent cloning of the resulting fragment. The vectors generated were designated pPBSProMLEVΨAkv-neo and pPBSGlnMLEVΨAkv-neo (see Fig. 2A).

Cells, transfections, and virus infections.

Growth conditions for Ψ-2 (34) and NIH 3T3 cells and transfections of packaging cells and selection for stably integrated vectors have been previously described (33, 38). Briefly, 10 μg of vector DNA was transfected into Ψ-2 packaging cells seeded at 5 × 10³ cells per cm² on the day before transfection. To estimate the level of marker gene expression in vectors with various 5′ UTR lengths, G418-resistant colonies were counted before pooling. Virus infection and determination of transductional efficiencies were carried out as previously described (38). Briefly, serially diluted virus-containing medium was transferred to NIH 3T3 cells; after 10 days of selection of recipient cells, resistant colonies were counted, individually isolated, and expanded. In some experiments, G418-resistant NIH 3T3 colonies on each plate were pooled to allow for a PCR-based screening among a large number of transduction events. In a transduction series set up for colony pooling, virus-containing medium was diluted to obtain approximately 10 G418-resistant colonies per plate.

Proviral DNA sequence analysis.

Genomic DNA from G418-resistant clones and colony pools was prepared as previously described (33). Sequence analysis of individual transduced vector sequences was performed on a PCR product encompassing part of the 5′ LTR, the PBS, the 5′ UTR, and the upstream part of the neo gene. The PCR was performed with oligonucleotide 1 (ON1), matching Akv MLV positions 7838 to 7865 (64) (5′-TTCATAAGGCTTAGCCAGCTAACTGCAG-3′), and ON2, matching neo positions 1656 to 1683 (3) (5′-GGCGCCCCTGCGCTGACAGCCGGAACAC-3′). The resulting PCR product was sequenced by use of an upstream primer (ON3) matching Akv MLV positions 69 to 96 (64) (5′-TCCGAATCGTGGTCTCGCTGATCCTTGG-3′) and, for relevant clones harboring PBS-Gln, by using a downstream primer (ON4) matching neo positions 1223 to 1244 (2) (5′-CTTCCTTTAGCAGCCCTTGCGC-3′).

PCR-based screening for recombinants.

A PCR-based screening for Akv-MLEV recombination within the 5′ UTR was performed on genomic DNA prepared from colony pools each obtained by pooling of all G418-resistant colonies (approximately 10 colonies per plate) obtained on a single plate. In the PCR amplification, a primer matching the MLEV PBS-Gln (ON5; 5′-GTCTTTCATTTGGAGGTCCCA-3′) and a neo-specific primer (ON2) were used. The resulting PCR product (if any) for each pool was sequenced with ON4 and ON5.

Nucleotide sequence accession number.

The MLEV sequence determined in this study has been assigned GenBank accession no. AF041383.

RESULTS

5′ UTR sequences of Akv-MLV and MLEV.

MLEV is encapsidated into virus particles released from Ψ-2 packaging cells (38). The functional MLEV PBS-Gln may be used for initiation of reverse transcription as an alternative to an impaired vector PBS, and endogenous MLEV RNA may thus serve as a recombination partner in rescue of PBS-modified vectors (38, 39). To further characterize MLEV, we performed, among different PCR-based strategies used, a semirandom-primed PCR on cDNA synthesized from viral RNA from Ψ-2-derived virus particles. Resulting PCR products were sequenced with primers recognizing the PCR product termini, thereby obtaining sequences of the entire 5′ UTR and part of the gag coding region. Alignment of the 5′ UTRs of Akv and MLEV demonstrated homology throughout the region (Fig. 1). A total of 101 nucleotide position differences were found dispersed between the PBS and the gag start codon; these nucleotide differences were grouped into 34 leader markers (LM1 to LM34) representing single-nucleotide differences (e.g., LM3), clusters of differences (e.g., LM24), and deletions (e.g., LM5) or insertions (e.g., LM20) in MLEV. This genetic marker-based division of the 5′ UTR defines an array of SIWs (designated SIWI to SIWXX) ranging in size from 6 to 27 nucleotides (Fig. 1). MLEV harbors the glyco-gag and gag start codons at positions 375 and 639, respectively (Fig. 1). It is uncertain, however, whether functional MLEV glyco-Gag and Gag proteins are produced. Previous studies thus suggest that functional glyco-Gag protein is not encoded by the 5′ UTR of an endogenous virus closely related to MLEV (13).

FIG. 1.

Alignment of Akv and MLEV 5′ UTRs. The sequence of MLEV was determined by various PCR-based methods as described previously (38) and in the text. PBS sequences are indicated in bold letters. Identical nucleotide positions are indicated by asterisks; nucleotide insertions in Akv and MLEV are indicated by introduction of colons (:) in MLEV and Akv sequences, respectively. Single-nucleotide differences and clusters of differences are underlined in the MLEV sequence, and the marker number (LM1 to LM34) is given below each genetic marker within the packaging region ranging from the 3′ PBS (position 163) to the gag start codon (position 638); LM1 to LM16 correspond to markers IV to XVIII in reference 38. SIWs between markers are indicated by brackets and the designations SIWI to SIWXX. Glyco-gag (positions 375 to 377) and gag (positions 639 to 641) start codons are given in italics. Palindromes (eight nucleotides or longer) in Akv are underlined. The lengths of Akv and MLEV 5′ UTR sequences are 476 and 465 bp, respectively; due to a total of five nucleotide insertions in MLEV (indicated by colons in Akv sequence), the length of the entire recombination window is 481 bp.

Chemical modification studies of Moloney MLV RNA have suggested a highly ordered secondary structure of the 5′ UTR (1, 42, 62). The relevant regions of Moloney MLV, Akv MLV, and MLEV were put through a computer-based analysis (performed by use of RNAdraw [37]). On this basis, we predict that similar structure models may account for MLEV and Akv (indicated schematically in Fig. 2A); the only exception is the apparent lack of a stable stem-loop 6 (SL 6) in MLEV. According to the proposed RNA secondary structure model, four potential stem-loops are included in the vectors (pPBSPro244ΨAkv-neo and pPBSUmu244ΨAkv-neo [Fig. 2A]) which we have previously studied in replication and recombination experiments (38). These stem-loops include SL 2, involved in RNA dimerization, and SLs 3 and 4, required for RNA encapsidation (41, 43, 67). The role of SL 1 is unknown, as are the roles of putative SLs 5 to 9. SL 10 is part of an internal ribosome entry site involved in gag polyprotein translation (5, 63). SLs 5 to 9 and a putative shorter form of SL 10 have been included in vectors pPBSPro476ΨAkv-neo and pPBSUmu476ΨAkv-neo (Fig. 2A) tested in this study.

We conclude that Akv and MLEV 5′ UTRs are closely related and likely share a highly ordered secondary structure. The sequences differ only at scattered nucleotide genetic marker positions dispersed throughout the region. Template switching within the 5′ UTR of Akv and MLEV templates during minus-strand synthesis therefore represents an in vivo model system with which to evaluate whether homologous recombination between naturally occurring retroviral sequences is affected by the presence of SIWs, palindromic sequences, and putative stem-loop structures.

Transfer of vectors with various lengths of sequences between PBS and neo.

The versatile function of the MLV 5′ UTR suggests that differences in this region may influence both early and late events of retrovirus replication. To estimate whether differences between Akv and MLEV within the 5′ UTR affect the production of a protein encoded downstream from the region, the MLEV 5′ UTR was introduced into an Akv vector. The resulting vectors harboring PBS-Pro and PBS-Gln were designated pPBSProMLEVΨAkv-neo and pPBSGlnMLEVΨAkv-neo, respectively (Fig. 2A).

neo expression was assessed for vectors harboring PBS-Pro or PBS-Gln by transfection into Ψ-2 packaging cells and subsequent G418 selection. Thus, the number of G418-resistant colonies per transfection indicated whether neo expression was affected by the presence of the 476-bp Akv 5′ UTR or the 465-bp MLEV 5′ UTR compared, for example, to vectors harboring the shorter 5′ UTR. We did not detect any difference for vectors harboring complete Akv and MLEV 5′ UTR sequences (Fig. 2A). A higher level, however, was seen for the vector with the shorter version of the 5′ UTR.

To obtain an overall estimate of the effect of altering the PBS and the length of the 5′ UTR, we measured vector replication efficiencies in transductional titer assays. As expected, transduction of vectors harboring PBS-Umu was strongly diminished, with titer reductions of about 5 orders of magnitude compared to the PBS-Pro constructs (Fig. 2A). There was no significant difference in titer values obtained with pPBSUmu244Ψ and pPBSUmu476Ψ constructs. For vectors harboring the wild-type PBS sequence, however, we detected about a 10-fold increase in titer when the longer 5′ UTR was included (Fig. 2A).

Since the expression level is not influenced significantly by differences within Akv and MLEV 5′ UTRs, recombination between Akv-derived vectors harboring the 476-bp Akv 5′ UTR and endogenous MLEV may be studied without a bias for specific recombination sites due to selection at the level of marker gene expression. Furthermore, our transduction data may indicate that one or more cis-acting elements in the downstream part of the 5′ UTR are directly involved in retroviral replication or that an overall structure of the entire region is supportive for the actions of the stem-loops in the upstream part of the 5′ UTR.

Recombinational repair of PBS-impaired vectors harboring 476-bp 5′ UTR.

Previous results with PBS-impaired vectors harboring a shortened 5′ UTR indicate that the majority of the rare events of transduction of these vectors are mediated by recombination with MLEV RNA, thereby introducing PBS-Gln in the transduced provirus (38). In the present study of PBS-impaired vectors with the longer 5′ UTR, we initially tested by sequence analysis the PBS composition in 38 G418-resistant colonies, each representing an individual transductional event (Table 1). Surprisingly, we found that only two clones harbored PBS-Gln, suggesting a relatively low incidence of 5′ UTR minus-strand recombination with MLEV. The sequences of the proviral 5′ UTR in these two clones (28 and 42) are shown in Fig. 3. The remaining proviruses were results of 5′ UTR minus-strand recombination with the Moloney MLV-based packaging construct (34), R-U5-mediated second-strand transfer recombination involving MLEV (39) or the packaging construct (40), or transfer through yet unknown transduction pathways. None of the 25 proviruses that were transduced through unknown pathways harbored sequences of MLEV or packaging construct origin in upstream or downstream LTRs or in the 5′ UTR, indicating that these proviruses were generated from vector RNA homodimers by aberrant reverse transcription mechanisms.

TABLE 1.

neo transduction by PBS-Umu vectors harboring the Akv 476-bp 5′ UTR

Transduced PBS	No. of clones with specific PBS/38 pro- viruses analyzed	Transduction pathway
		5′ UTR minus-strand recombination with indicated recombi- nation partnerb		R-U5-mediated second-strand transfer recombinationa				Unknownd
				Recombination partner		Minus-strand DNA transferc
		MLEV	Mo-Ψ−	MLEV	Mo-Ψ−	Inter	Intra
PBS-Gln	2	2/2				2/2	0/2
PBS-Pro	2		2/2			2/2	0/2
				4/33		2/4	2/4
PBS-Umu	33				4/33	4/4	0/4
								25/33
ΔPBS	1							1/1e

^aTransduction was mediated through initial priming on MLEV RNA or packaging construct RNA (40), read-through of PBS-Umu during minus-strand synthesis, and subsequent second-strand transfer mediated by R-U5 complementarity of the plus strand and the extended minus-strand DNA acceptor template (39). 

^bTransduced PBS-Gln originates from recombination with MLEV; transduced proviral sequences 28 and 42 are given in Fig. 3. Transduced PBS-Pro originates from recombination with the Moloney MLV-derived Ψ-2 packaging construct (Mo-Ψ⁻), which harbors a 350-bp deletion within the packaging signal (34). Sequence identities were 43 of 48 positions upstream from deletion and 28 of 52 downstream from the deletion. Recombination sites were mapped within the upstream region (40). 

^cThe character of minus-strand DNA strong-stop transfer to the vector 3′ end (inter) or MLEV 3′ end (intra) was determined through sequence analysis of the U3 region in the downstream LTR. The presence of MLEV U3 sequences thus indicated intrastrand transfer. 

^dProviruses did not contain sequences of MLEV or packaging construct origin in the regions analyzed (5′ LTR, 5′ UTR, and 3′ LTR). 

^ePBS potentially deleted due to aberrant second-strand transfer (53). 

FIG. 3.

Nucleotide sequences of PBS-Gln-harboring transduced proviruses and recombination partners Akv and MLEV. The sequences of individual transduced proviruses were determined by sequence analysis of PCR fragments encompassing the entire 5′ UTR (here defined as the region from PBS to gag start codon). Transduced viral sequences are compared with homologous regions of Akv (top) and MLEV (bottom). Two sequences, 28 and 42, originate from analysis of individual colony clones (Table 1); the remaining sequences originate from PCR screening of colony pools obtained from separate plates and subsequent sequence analysis. Nucleotides homologous to positions in Akv are indicated by hyphens; deleted nucleotides compared to Akv are indicated by colons, whereas insertions are indicated by introduction of a colon in the Akv sequence. Genetic markers consisting of more than one-nucleotide differences are underlined. Molecular differences between Akv-neo and MLEV within the 5′ UTR are designated LM1 through LM34; marker numbering is indicated below the Akv sequence. LM1 to LM16 correspond to markers IV to XVIII in reference 38. The gag start codon is indicated for convenience (position 639); however, the ATG sequence was not included in the vectors utilized. R, A/G mixed nucleotide position.

Most likely, the capacity to form Akv-MLEV heterodimers allowing for recombination is disturbed by the longer 5′ UTR. This observation is consistent with the higher titers observed for vectors harboring the 476-bp 5′ UTR and possibly reflects that vector RNAs with the entire UTR are more likely to generate packageable homodimers. Alternatively, pPBSUmu476Ψ vectors are generated at a higher level than vectors with a 244-bp 5′ UTR in G418-selected packaging cells and therefore may diminish heterodimerization with MLEV, resulting in a lower incidence of vector-MLEV recombination.

PCR screening for recombination with MLEV.

To reveal additional template switching events within the 5′ UTR, a large number of colonies were screened for Akv-MLEV recombinant proviruses harboring PBS-Gln. Twenty-five colony pools were generated by pooling of G418-resistant NIH 3T3 colonies obtained on 25 separate plates, each containing approximately 10 G418-resistant colonies. Colony pools were screened by PCR with primers matching specifically PBS-Gln and sequences within the neo gene. PCR products were obtained in 21 of 25 amplifications, and the resulting DNA fragments were sequenced with PBS-Gln and neo primers (Fig. 3). This strategy presented the possibility that two or more proviruses could have been simultaneously amplified in the same PCR. However, in only one case (pool O), two distinct PBS-Gln-containing proviral sequences were evident in the product amplified from the same pool. This resulted in the generation of an A/G mixture position at LM9. Obviously, we cannot exclude the possibility that two identical proviruses were amplified from the same pool.

Sequence analysis of PBS-Gln-containing proviruses revealed the specific MLEV pattern of nucleotide differences from Akv flanked downstream by Akv sequences (Fig. 3). Sequences ranged from harboring only the PBS-Gln of MLEV origin (pools S and V) to harboring markers LM1 to LM28 of MLEV origin (pools F and J). We conclude that PBS-impaired vectors with a complete 5′ UTR undergo recombinational repair through initiation of reverse transcription on the functional MLEV PBS and subsequent minus-strand template switching within the 5′ UTR to obtain the PBS complementarity required for efficient plus-strand transfer (Fig. 4). The genetic markers dispersed throughout the region allowed us to map specifically the site of template switching in each transduction event. Recombination sites were clustered within markers LM8 and LM10. Thus, we mapped 5 template shifts within SIWVIII and 11 within SIWIX, whereas template switching occurred within SIWI, SIWIII, and SIWXIX in two, four, and two cases, respectively (Fig. 4).

FIG. 4.

Mapping of sites for recombinational repair of PBS-modified vectors. Template switching during minus-strand DNA synthesis within the 5′ UTR between the PBS and neo gene is established to obtain perfect PBS complementarity in second-strand transfer. Thin lines indicate RNA; thick lines represent DNA. The sloping line represents the 5′ UTR with genetic markers LM1 to LM34. Stippled dots indicate single-nucleotide differences, whereas clusters of differences are indicated by black dots. Lengths of SIWs are given between genetic markers. Recombination sites are dispersed throughout the region as delineated by arrow width and ratios (number of proviruses with specific recombination site/total number of proviruses analyzed). In 16 of 24 analyzed proviral sequences harboring PBS-Gln, the recombination site was mapped within a region coinciding with the kissing stem-loop dimerization domain. The kissing stem-loop (SL 2 [Fig. 2A]) harbors LM8 and LM9, whereas LM10 is located within SL 3 (Fig. 2A).

We did not in any case register sequence abnormalities such as deletions, insertions, or single-base mutations in the proviral recombinants analyzed (Fig. 3). Misincorporations were thus not evident at the point of transfer of the growing DNA strand, and we propose that in vivo reverse transcriptase-mediated recombination between naturally occurring retroviral sequences is not an error-prone process.

In conclusion, precise recombination sites were mapped and found clustered within specific regions of the Akv-MLEV 5′ UTR. In addition, we did not in any proviral recombinant find evidence for three template shifts within the 5′ UTR.

Recombination within Akv-MLEV SIWs.

Having identified the exact recombination sites in 24 distinct events involving MLEV, we sought to deduce whether specific sites of template switching were preferred by the reverse transcriptase. Among the 20 SIWs (defined as DNA stretches with six or more identical nucleotides) which appeared from the alignment of Akv and MLEV 5′ UTRs (Fig. 1), we found recombination sites clustered within 5 (Table 2). This pattern of recombination sites did not reflect a selective bias at the level of transcription or translation of the proviral neo gene, as indicated by the similar levels of marker gene expression obtained with vectors harboring the Akv and MLEV 5′ UTRs (Fig. 2A). We thus posit that downstream genetic markers of MLEV origin do not interfere with marker gene expression, and therefore selection at the level of gene expression does not result in discrimination of any sites of template switching. We conclude that the recombination sites detected were not randomly distributed throughout the 5′ UTR but that some windows were more prone to template switching than others.

TABLE 2.

Recombination within Akv-MLEV sequence identity windows

SIW	Length (nta)	Location	Recombination within window (recombinants/total)	Sequential and structural characteristics
Single
I	11	PBS-LM1	2/24
II	19	LM1-LM2
III	21	LM2-LM3	4/24	10-nt palindrome
IV	6	LM3-LM4
V	16	LM4-LM5
VI	14	LM6-LM7		8-nt palindrome
VII	12	LM7-LM8
VIII	12	LM8-LM9	5/24	Upstream leg of kissing stem-loop, SL 2 (and 6-nt palindrome)
IX	24	LM9-LM10	11/24	Downstream leg of kissing stem-loop, SL 2 (and 6-nt palindrome)
X	27	LM11-LM12		SL 3
XI	8	LM12-LM13		Upstream leg of SL 4
XII	12	LM13-LM14		Loop, SL 4 (and 10-nt palindrome)
XIII	11	LM16-LM17		Upstream leg of SL 5 (and 6-nt palindrome)
XIV	8	LM17-LM18		Downstream leg of SL 5
XV	7	LM22-LM23		Downstream leg of SL 6
XVI	13	LM25-LM26		Downstream leg of SL 7
XVII	8	LM26-LM27		Upstream leg of SL 8
XVIII	23	LM28-LM29		Upstream leg of SL 9
XIX	23	LM29-LM30	2/24	Downstream leg of SL 9 (and 6-nt palindrome)
XX	6	LM31-LM32
Doubleb
I + II	31	PBS-LM2	2/24
III + IV	28	LM2-LM4	4/24	10-nt palindrome
VIII + IX	37	LM8-LM10	16/24	Kissing loop, SL 2 (and 16-nt palindrome)
X + XI	36	LM11-LM13		Part of SL 4
XVIII + XIX	47	LM28-LM30	2/24	SL 9 (and 6-nt palindrome)

^ant, nucleotides. 

^bWindows covering two adjacent SIWs separated by one single-nucleotide difference. 

Recombination-prone windows ranged in size from 11 to 24 nucleotides (Table 2); thus, recombination was not detected within windows smaller than 11 nucleotides. In case of random site distribution, we would expect to observe template shifting also within the smaller SIWs. The possibility exists that a certain length of sequence homology or sequence identity is required for strand transfer. However, the fact that large SIWs (such as SIWII, SIWX, and SIWXVIII [Table 2]) did not harbor any recombination site suggested that donor-acceptor homology is not the primary determinant in template switching. As delineated in Table 2, there is no direct correlation between length of SIW and number of sites. Considering the five largest SIWs defined by allowing one single-nucleotide difference (Table 2), we likewise deduce that recombination is not a simple matter of sequence similarity for the reverse transcriptase to shift template. As a striking example, two neighboring windows (SIWVIII-IX and SIWX- XI) of similar lengths (37 and 36 nucleotides, respectively) appeared to support recombination in 16 (66%) and no transductional events, respectively. However notably, our data do not exclude that sequence homology may be a prerequisite for efficient template switching.

DISCUSSION

Forced recombination.

We have previously seen that PBS knockout Akv MLV vectors may restore function (38) or be rescued (39) through recombination with MLEV, a retroviral sequence endogenous to the murine host genome. In this study, we used impairment of the Akv PBS to study specifically by forced recombination Akv-MLEV heterodimerization and recombination between retroviral species with distinct but homologous 5′ UTR sequences. To study template switching events within the leader region which harbors multiple cis-acting viral elements, we exploit the dual role of the PBS in initiation of and strand transfer during proviral DNA synthesis. Hence, cDNA synthesis initiation on copackaged MLEV RNA and transfer of minus-strand strong-stop DNA to the 3′ end of the vector result in subsequent copying of the MLEV PBS-Gln during plus-strand strong-stop DNA synthesis. Consequently, conventional minus-strand synthesis without strand exchange leads to a dead end due to the lack of complementarity between the mutated PBS and PBS-Gln. Second-strand transfer may in this situation be mediated by complementary sequences within R and U5 regions (39). However, another possibility is switching of the nascent minus-strand DNA from the Akv donor to the MLEV acceptor RNA template within the 5′ UTR to obtain perfect PBS-Gln complementarity during plus-strand transfer. Therefore, a transduced PBS-Gln sequence is indicative of the occurrence of recombination within the 5′ UTR and may, as in this study, be used for screening among a large number of transduction events.

Clustering of recombination sites and template shift precision.

In this work, we analyzed 24 independent events of template switching within the complete MLV 5′ UTR. Nucleotide differences between the recombination partners allowed us to map the positions where the proceeding reverse transcriptase transfers from donor to acceptor template. Recombination sites were distributed nonrandomly throughout the 481-nucleotide 5′ UTR recombination window. This apparent clustering of template switching sites was not caused by a selective bias at the level of marker gene expression, as demonstrated by the ability of MLEV 5′ UTR-containing vectors to confer G418 resistance upon murine fibroblasts. Also, this observation is supported by the fact that in two proviruses recombination had occurred near the 3′ end of the 5′ UTR and by previous studies showing that MLEV-related leader sequences do not interfere with expression of downstream genes (13, 21).

Recombination was precise in all recombinants analyzed and thus did not result in any nucleotide misincorporations at frequent or less frequent transfer sites. This observation is consistent with our previous observations of exact template switching within the 5′ UTR (38) and with observations by Zhang and Temin demonstrating that homologous recombination in vivo is not error prone (68). Discrepancies with studies showing that junction sites harbor misincorporations in vitro (51, 66) may be due to the fact that in vitro studies do not reflect template switching in vivo or that distinct mechanisms for retroviral recombination differ in their ability to confer precise template switching. The process of nascent-strand transfer may therefore not necessarily in itself contribute to retroviral diversity.

Recombination governed by template homology?

The nonrandom pattern of recombination potentially resides in a certain degree of homology between the two templates or in primary or RNA secondary structures within the region. Studies by Zhang and Temin (68) indicate that recombination between nonhomologous RNA templates depends on the length of an inserted sequence identity region. However, a detailed look at the entire region (Fig. 1) and its division into SIWs (Table 2) suggests that homology and template similarity in this case are not the primary determinants for retroviral recombination. If this were so, we would expect recombination sites to be more evenly distributed throughout the region due to extensive homology between the two recombination partners. Since the preferred region of recombination within SIWVIII and SIWIX coincides with the region previously found to mediate recombination in shorter vectors (38), we may also conclude that a need for a minimum length of homology between the growing DNA strand and the RNA acceptor template is not the decisive factor in template shifting. If this were the case, we would observe a substantial amount of recombination events in the downstream part of the 5′ UTR, likely within SIWXVIII+XIX.

Palindromic sequences involved in recombination?

Palindromic sequences may in theory represent local RNA-RNA interaction sites possibly contributing to kissing-loop-mediated dimerization. This notion may be supported by the fact that dimeric RNA interacts at multiple sites along the genome (4). The presence of 16- and 10-nucleotide-long palindromic sequences within SIWXVIII+IX (hosting 16 of 24 sites of recombination) and SIWIII (hosting 4 of 24 sites of recombination), respectively, may indicate some significance of palindromic sequences in template switching and possibly dimerization. However, 8- and 10-nucleotide palindromes within SIWVI and SIWXII, respectively, did not in any case promote template switching. Therefore, we cannot from our analysis of recombination sites determine whether template shifts are facilitated by the presence of palindromes in the RNA templates.

Site-specific recombination due to RNA secondary structures?

Secondary structures may pause the reverse transcriptase during minus-strand synthesis (16); this pausing promotes strand transfer in vitro (6, 17, 66). In our experiments, 5 of 16 recombination events in the kissing stem-loop were mapped within the upstream leg of the stem-loop (SIWVIII [Table 2]). In addition, other secondary structures, such as the well-established double stem-loop structure (SLs 3 and 4 [Fig. 2]) (43, 67) downstream from the dimerization region, did not promote template shifting. In contrast to our previous studies of vectors containing SLs 1 to 4, we can from the present work state that the reverse transcriptase pausing and sequence homology combined do not promote template switching in the region downstream from the stable SL 4. In summary, we do not believe that pausing of the reverse transcriptase due to intramolecular RNA structures contributes significantly to the recombination process. A similar conclusion was drawn in a recent in vitro study of internal strand transfer at the HIV-1 transactivation response region situated in the R region. In this case, the structure-derived pause site did not coincide with the preferred transfer site, and it was suggested that the close interaction between donor and acceptor stem-loops is the driving force for template switching (28).

Based on our results and the considerations presented above, we propose that a specific RNA structure is crucial for recombination within the MLV 5′ UTR. The preferred region of template shifting thus coincides precisely with the kissing stem-loop demonstrated by in vitro studies to be crucial for MLV RNA dimerization (20, 52, 62). Considerable evidence derived from both in vitro and in vivo experiments indicates that the corresponding stem-loop in HIV-1 is involved in, but possibly not essential for, dimerization (4, 8, 9, 14, 22, 30, 31, 35, 44, 47–49, 57). This colocalization of the hot spot recombination site and the primary dimerization domain suggests a relevant connection between the processes of dimerization and recombination. Thus, template shifting at this particular site may be favored by the proximity of RNA templates and by direct RNA-RNA interactions in this part of the genome. It is possible, however, that the palindromic character of the region, allowing perfect acceptor template match perhaps combined with pausing of the reverse transcriptase when it encounters and traverses the stable intermolecular RNA structure at this site, may support strand transfer promoted by the close proximity of the interacting RNA molecules. Also, frequent RNA breakage within certain secondary structures may promote forced copy-choice recombination (10).

Model for kissing-loop-mediated retroviral recombination.

In summary, we favor the idea that intimate RNA-RNA interactions due to kissing-loop-mediated dimerization are the driving force for recombination within the 481-bp 5′ UTR recombination window. Dissociation of the reverse transcriptase from the donor template within this region may thus lead to a shift of template and continued DNA synthesis on the acceptor template. Recent studies suggest that the reverse transcriptase may otherwise tend to reassociate with the donor template RNA (25). Our data also indicate that heterodimerization of Akv and MLEV is mediated by kissing-loop interactions, despite the presence of a single-nucleotide difference within the six-nucleotide loop sequence. Noteworthy, perfect base pairing is restored by G (MLEV):U (Akv) base pairing, and the primary interaction may therefore not be disturbed. As illustrated in Fig. 5, we propose that the reverse transcriptase traversing the RNA dimerization region switches template due to template proximity and possibly reverse transcriptase pausing within or near the putative RNA duplex formed. It should be emphasized that we cannot in the present experimental setup determine the exact frequency of template switching within the particular region and therefore do not know whether the kissing-loop may act as a frequent mediator of recombination also in retroviral replication devoid of selective constraints.

FIG. 5.

Model for kissing-loop-mediated recombination (not drawn to scale). The interaction of Akv and MLEV kissing loops and putative subsequent RNA duplex formation promote template switching during viral DNA synthesis, most likely due to close RNA-RNA interactions in the region. Enlargements of the interacting Akv and MLEV stem-loops and the RNA duplex supposedly generated subsequent to loop kissing are shown. It should be noted that the interaction of kissing loops represents a local antiparallel linkage; for convenience, the two interacting RNA templates are presented in antiparallel orientation.

This model for kissing-loop-mediated recombination suggests that the kissing-loop interaction, and possibly RNA duplex formation, is intact also within the nucleocapsid core particle during reverse transcription of the mature RNA dimer. This may indicate that linkage in this particular area of the dimeric genome is crucial for the overall structure of the dimer and perhaps strand transfer during viral DNA synthesis, a notion supported by the fact that deletion of the dimerization stem-loop severely inhibits plus-strand DNA transfer (47). Consistent with this idea, it was recently proposed that close RNA interactions possibly within the dimer linkage site of the MLV packaging sequence may promote template switching during reverse transcription of a downstream gene (15). However, the view of a persistent kissing-loop palindrome interaction is challenged by the observation that kissing-loop mutations in HIV-1 do not influence RNA dimer stability, indicating, according to the authors, that the initial kissing-loop interaction is resolved during maturation of the dimer (4). At this stage, we cannot explain these discrepancies.

Similarity to RNA structure-based template switching in other RNA viruses.

The complementing action of a high mutation rate and frequent genetic recombination is a key player in maintenance of viability and variability in diverse groups of RNA viruses. Evidence suggests that related mechanisms may account for template switching during viral RNA or DNA synthesis in diverse viruses. Results obtained from studies of picornavirus (poliovirus) (54, 61), bromovirus (brome mosaic virus) (46), and now mammalian type C retrovirus (MLV) (this study) recombinants indicate that local regions of hybridization between identical sequences are preferred sites for template switching. In addition, secondary RNA structures have been found to be crucial for recombination in various RNA viral species. Hence, in plant viruses such as brome mosaic virus (46), turnip crinkle virus (7), and tombusviruses (65) and in animal viruses such as picornaviruses (e.g., human poliovirus) (54) and coronavirus (mouse hepatitis virus) (55), stem-loop structures are primary mediators of template switching. Our results provide another example of viral recombination mediated by defined structural elements in the genome, thus adding support to the notion that recombination mechanisms crucial for viral evolution are exploited by a diverse spectrum of viral species. Common to all of these studies is that preferred recombination sites coincide with regions of predicted RNA secondary structure. Although the exact mechanism in most cases remains a matter of speculation, current knowledge indicates that stem-loop structures contribute to site-specific recombination by mediating intermolecular template interactions or by promoting polymerase pausing. The results presented in this report shed light on recombination at the primary site of viral RNA interaction and thus raise the possibility that alternative interaction sites in the RNA genome play similar roles in retrovirus replication and evolution.