Intermolecular V(D)J Recombination*

V(D)J recombination plays a prominent role in the generation of the antigen receptor repertoires of B and T lymphocytes. It is also likely to be involved in the formation of chromosomal translocations, some of which may result from interchromosomal recombination. We have investigated the potential of the V(D)J recombination machinery to perform intermolecular recombination between two plasmids, either unlinked or linked by catenation. In either case, recombination occurs in trans to yield signal and coding joints, and the results do not support the existence of a mechanistic block to the formation of coding joints in trans . Instead, we observe that linearization of the substrate, which does not alter the cis or trans status of the recombination signals, causes a specific and dramatic reduction in coding joint formation. This unexpected result leads us to propose a “release and recapture” model for V(D)J recombination in which coding ends are frequently released from the postcleavage complex and the efficiency of coding joint formation is influenced by the efficiency with which such ends are recaptured by the complex. This implies the existence of mechanisms, operative during recombination of chromosomal substrates, that act to prevent coding end release or to facilitate coding end recapture. V(D)J recombination is the process of assembly of T-cell and immunoglobulin genes from V, J, and sometimes D gene segments (1). Lymphoid-specific proteins RAG1 (recom-bination activating gene 1) and RAG2 (2, 3) and multiple ubiq-uitously expressed protein factors are essential for V(D)J recombination. RAG1 and RAG2 bind to and cleave at specific 0.5 m M dithiothreitol, and 50% glycerol. Samples not requiring catenation were treated in an identical manner, except that gd -resolvase was substituted with its storage buffer. Samples were then digested with Xcm I or Afl III in an appropriate buffer (New England Biolabs) for 1 h at 37 °C and 10 min at 70 °C. If the samples did not require restriction digestion at this point, they were treated in an identical manner except that restriction endonuclease was substituted with 50% glycerol. To perform RAG-mediated cleavage, DNA samples (200 ng, 45 fmol) were incubated with 80 ng of RAG1 and 80 ng of RAG2 purified from F2A1 cells, and 0.3 ng/ m l of murine HMG2 (high mobility group 2; expressed and purified from E. coli ) (28) in 20 m M HEPES-NaOH, pH 7.5, 10 m M magnesium acetate, 50 m M sodium acetate, 10 m M ZnSO 4 , 2 m M dithiothreitol, 5% glycerol, and 0.1 mg/ml acetylated bovine serum albumin (total volume, 100 m l) for 3 h at 37 °C. In thenegative control, RAG1 and RAG2 were substituted with 50% glycerol. After cleavage, Tris-HCl, pH 8.3, was then added to 75 m M , EDTA to 7.5 m M , SDS to 0.15%, and proteinase K to 0.2 mg/ml, and samples were incubated fo r 1 h at 55 °C. 10 m g of yeast tRNA was added as a carrier, NaCl was added to 200 m M , and the samples were ethanol precipitated. They were resuspended in TE buffer (10 m M Tris-HCl, pH 8.0, 1 m M EDTA) containing 10 m g/ml of RNaseA, and samples that had not been treated with a restriction endonuclease were treated with it at this point. Half of the DNA was resolved on a 0.8% agarose gel, transferred to a GeneScreen Plus membrane (NEN Life Science Products), and hybridized and AT2c-r primers. PCR conditions were as described above except that 28 cycles were performed. In all PCR reactions using the Mark1 primer, an annealing temperature of 58 °C was used. Products of PCR reactions (3 m l) were resolved on 5% native polyacrylamide gels, transferred to GeneScreen Plus membranes, hybridized with 32 P-labeled SJ2 oligonucleotide (3) (5 9 -ctg tgc aca gtg gta) with a 50 °C wash with SSC for signal joints or with AT2-5 (5 9 -ggc tgc agg iit gta tcc taa where inosine) with with for and residues in 32 P-labeled oligonucleotide, PCR Clones selected random without prescreening by hybridization for sequencing.

V(D)J recombination plays a prominent role in the generation of the antigen receptor repertoires of B and T lymphocytes. It is also likely to be involved in the formation of chromosomal translocations, some of which may result from interchromosomal recombination. We have investigated the potential of the V(D)J recombination machinery to perform intermolecular recombination between two plasmids, either unlinked or linked by catenation. In either case, recombination occurs in trans to yield signal and coding joints, and the results do not support the existence of a mechanistic block to the formation of coding joints in trans. Instead, we observe that linearization of the substrate, which does not alter the cis or trans status of the recombination signals, causes a specific and dramatic reduction in coding joint formation. This unexpected result leads us to propose a "release and recapture" model for V(D)J recombination in which coding ends are frequently released from the postcleavage complex and the efficiency of coding joint formation is influenced by the efficiency with which such ends are recaptured by the complex. This implies the existence of mechanisms, operative during recombination of chromosomal substrates, that act to prevent coding end release or to facilitate coding end recapture.
V(D)J recombination is the process of assembly of T-cell receptor and immunoglobulin genes from V, J, and sometimes D gene segments (1). Lymphoid-specific proteins RAG1 (recombination activating gene 1) and RAG2 (2,3) and multiple ubiquitously expressed protein factors are essential for V(D)J recombination. RAG1 and RAG2 bind to and cleave at specific recombination signal sequences (RSSs) 1 generating blunt signal ends and covalently sealed, hairpin coding ends (4 -7). RSSs consist of conserved heptamer and nonamer sequences separated by a nonconserved spacer 12 or 23 base pairs long . V(D)J recombination occurs efficiently in vivo only between RSSs with different spacer lengths, a restriction known as the 12/23 rule (8).
Recent biochemical studies indicate that V(D)J recombination proceeds through a series of protein-DNA complexes. Prior to cleavage, the RAG1 and RAG2 proteins form stable com-plexes with individual and synapsed pairs of RSSs (9 -17). After cleavage, the RAG proteins remain tightly bound to a synapsed pair of signal ends (15,18) and also appear to interact with the two hairpin coding ends, albeit with lower affinity (15). Therefore, the immediate product of cleavage is thought to be a "cleaved signal complex" containing the four free ends (15).
Usually, RSSs involved in V(D)J recombination are located on the same chromosome, i.e. in cis. However, V(D)J recombination has been implicated in the formation of chromosomal translocations leading to lymphoid malignancies (19 -23). In certain cases, such translocations could be the result of V(D)J recombination occurring in trans, using RSSs on different chromosomes. There have been attempts to address this issue. The first such study attempted, and failed, to detect V(D)J recombination in trans between two plasmids in a transient transfection assay (24). However, the assay used may not have been sufficiently sensitive to detect the expected level of trans recombination (see "Discussion"). More recently, it was shown that at relatively high concentrations, oligonucleotides separately containing 12-and 23-RSSs can synapse and undergo coupled cleavage in vitro (15). Furthermore, interchromosomal V(D)J recombination has been detected in Abelson virus-transformed pre-B cells (25). Finally, it was reported that 12-and 23-RSSs on separate transiently transfected plasmids could mediate relatively efficient synapsis, coupled cleavage, and signal joint formation. This study also reported a specific and dramatic defect in coding joint formation if the two RSSs were in trans, which led to the proposal that mechanisms exist that selectively disfavor or inhibit coding joint formation in trans (26).
In contrast, we find that both coding and signal joint formation can occur relatively efficiently in trans. Our results do not support the existence of the mechanistic block to trans coding joints formation. Instead, a series of experiments with circular, catenated, or linear substrates suggest a model in which the efficiency of coding joint formation is strongly influenced by the efficiency with which coding ends can be recaptured after being released from the post-cleavage complex. Our data suggest that in the transient transfection assay, the vast majority of coding joints form after at least one cycle of coding end release and recapture.

EXPERIMENTAL PROCEDURES
Cell Lines-The F2A1 cell line, expressing heat shock-inducible murine RAG1 (amino acids 264 -1008) and RAG2 (amino acids 1-387), each containing a C-terminal extension of nine histidines and three copies of the c-myc epitope tag, has been described previously (27,28). Dr3 cells express heat shock-inducible full-length murine RAG1 and RAG2 (29), whereas M12 is the parental cell line of F2A1 and Dr3 and does not express RAG1 or RAG2. Chinese hamster ovary (CHO) cells were grown in ␣-modified Eagle's medium supplemented with 2 mM L-glutamine, 100 units/ml penicillin, 100 g/ml streptomycin, 4 g/ml each of adenosine, deoxyadenosine, and thymidine, and 10% fetal bovine serum.
Plasmids-To construct pAT2, the ␥␦-resolvase site was amplified by PCR from pNG367 (30) (a gift from Dr. Nigel Grindley) with 5Ј-ccg agg * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18  gtt ttt gta cag cca gac agt gga gta cta cca ctg tgg gat ctg tat cct aaa tca aat atc g and 5Ј-cca cca gga tcc tgt ccg aaa tat tat aaa tta tc primers for five cycles and then with the latter primer and 5Ј-cca cca aga tct ccg agg gtt ttt gta cag c primer for 25 additional cycles and subcloned into the BamHI and BglII sites of pJH290 (31). The second ␥␦-resolvase site was subcloned into the AatII site after it was amplified by PCR from pNG367 using 5Ј-cca cca acc gac gtc tgt ccg aaa tat tat aaa tta tc and 5Ј-cca cca acc gac gtc tgt atc cta aat caa ata tcg primers. The distance between the 12-and the 23-RSSs in pAT2 was 0.4 kb. To construct pAT6, the ␥␦-resolvase site was amplified by PCR from pNG367 with 5Ј-cca cca gga tcc tgt ccg aaa tat tat aaa tta tc and 5Ј-cca cca aga tct ccg agg gat ctg tat cct aaa tca aat atc g primers and subcloned into the BamHI and BglII sites of pAT2, thus deleting the 12-RSS. To delete the 23-RSS and produce pAT7, pAT2 was digested with SalI and recircularized.
In Vitro Cleavage Assay-Plasmid substrates (50 ng/l) were treated with 1/10 volume of ␥␦-resolvase (a gift from Dr. Nigel Grindley) in 20 mM Tris-HCl, pH 8.3, and 10 mM MgCl 2 for 1 h at 37°C and then for 10 min at 70°C. The ␥␦-resolvase stock solution was 10 g/ml in 10 mM Tris-HCl, pH 7.5, 1 M NaCl, 0.1 mM EDTA, 0.5 mM dithiothreitol, and 50% glycerol. Samples not requiring catenation were treated in an identical manner, except that ␥␦-resolvase was substituted with its storage buffer. Samples were then digested with XcmI or AflIII in an appropriate buffer (New England Biolabs) for 1 h at 37°C and 10 min at 70°C. If the samples did not require restriction digestion at this point, they were treated in an identical manner except that restriction endonuclease was substituted with 50% glycerol. To perform RAGmediated cleavage, DNA samples (200 ng, 45 fmol) were incubated with 80 ng of RAG1 and 80 ng of RAG2 purified from F2A1 cells, and 0.3 ng/l of murine HMG2 (high mobility group 2; expressed and purified from E. coli) (28) in 20 mM HEPES-NaOH, pH 7.5, 10 mM magnesium acetate, 50 mM sodium acetate, 10 M ZnSO 4 , 2 mM dithiothreitol, 5% glycerol, and 0.1 mg/ml acetylated bovine serum albumin (total volume, 100 l) for 3 h at 37°C. In the negative control, RAG1 and RAG2 were substituted with 50% glycerol. After cleavage, Tris-HCl, pH 8.3, was then added to 75 mM, EDTA to 7.5 mM, SDS to 0.15%, and proteinase K to 0.2 mg/ml, and samples were incubated for 1 h at 55°C. 10 g of yeast tRNA was added as a carrier, NaCl was added to 200 mM, and the samples were ethanol precipitated. They were resuspended in TE buffer (10 mM Tris-HCl, pH 8.0, 1 mM EDTA) containing 10 g/ml of RNaseA, and samples that had not been treated with a restriction endonuclease were treated with it at this point. Half of the DNA was resolved on a 0.8% agarose gel, transferred to a GeneScreen Plus membrane (NEN Life Science Products), and hybridized to 32 P-labeled AT2-2 probe. AT2-2 was excised from pAT2 with ApaI and EcoO109I.
In Vivo Recombination-In experiments studying the effect of catenation, substrates were prepared as described above except that larger scale reactions with ␥␦-resolvase were used and no RAG treatment was performed. In the experiments studying the effect of relaxation and linearization, the substrates (pAT2, pAT6, or pAT7) were treated or mock-treated with human topoisomerase I (TopoGEN, Inc.) (0.5 unit/l; 100 ng/l DNA) in 10 mM Tris-HCl, pH 7.9, 150 mM NaCl, 1 mM EDTA, 0.1% bovine serum albumin, 0.1 mM spermidine, 5% glycerol, and treated or mock-treated with NcoI. All samples were treated with proteinase K, isopropanol precipitated, and quantitated by UV absorption.
M12, F2A1, and Dr3 cells were heat shocked 6 -10 h prior to electroporation for 6 min. at 42°C or not heat shocked (F2A1 and Dr3 have relatively high uninduced expression levels of RAG1 and RAG2). Cells (2.5 ϫ 10 6 ) were electroporated with 10 g of DNA. Transfection efficiency was determined by transfection with pCMV␤ (Invitrogen) and staining the cells for ␤-galactosidase. Because pCMV␤ transfection was done in parallel with other samples in a transfection experiment, ␤-galactosidase expression did not interfere with the rate of recombination. Possible differences in transfection efficiencies between the pCMV ␤-sample and the actual samples used to measure recombination rates were corrected for by calculating the ratio of coding joint to signal joint formation, thereby measuring the relative amounts of the two types of joints. CHO cells were transfected using FuGENE 6 (Roche Molecular Biochemicals), and the plasmids pR1-264 and pMS216, which transiently express the same versions of RAG1 and RAG2 as those stably expressed in F2A1, were co-transfected with the substrate plasmids.
Extrachromosomal DNA was recovered 14 -16 h after transfection using an alkaline lysis method and the Qiagen Spin Miniprep kit. DNA was quantitated by UV absorption and by PCR with 5Ј-cct tac tgg gta ttt gtc tc and 5Ј-tct agt gca aag tct gct ac primers complementary to a nonrearranging part of the plasmids for 23 cycles of 35 s at 95°C, 35 s at 53°C, and 35 s at 72°C (12 pg/l DNA). UV data were used to calculate amounts of DNA. Quantitation by PCR was always in agreement with the UV data. Because DNA recovery was equivalent for transfections involving linear and circular substrates (and most substrate DNA did not enter the cells), the alkaline lysis method did not result in the loss of linear DNA molecules.
Coding joints of catenated samples were detected by PCR with AT2a-f (5Ј-cta ttt tta tag gtt aat gtc atg) and Inne1 (5Ј-cga cag gtt tcc cga ctg g) primers; signal joints were detected with AT2b-r (5Ј-gaa aag tgc cac ctg acg tc) and AT2c-r (5Ј-cta tat ggc tga tgc aga cag) primers. PCR reactions were performed for 35 cycles of 35 s at 95°C, 35 s at 64°C, and 30 ϩ 1 s/cycle at 72°C with the initial incubation for 2 min at 95°C and the final extension of 7 min at 72°C. Each reaction (25 l) contained 2.5 mM MgCl 2 and 0.05 unit/l of Platinum Taq (Life Technologies, Inc.). Coding joints of noncatenated samples were detected with Stop2 (5Ј-gtg aga atc gca gca act tg) and Inne1 primers (32); signal joints were detected with Mark1(32) (5Ј-cct gcc gag tgg ttc agc) and AT2c-r primers. PCR conditions were as described above except that 28 cycles were performed. In all PCR reactions using the Mark1 primer, an annealing temperature of 58°C was used. Products of PCR reactions (3 l) were resolved on 5% native polyacrylamide gels, transferred to GeneScreen Plus membranes, hybridized with 32 P-labeled SJ2 oligonucleotide (3) (5Ј-ctg tgc aca gtg gta) with a 50°C wash with 2ϫ SSC for signal joints or with AT2-5 (5Ј-ggc tgc agg iit gta tcc taa a, where i ϭ inosine) with a 55°C wash with 2ϫ SSC for coding joints, and quantitated on a Phos-phorImager. The AT2-4 oligonucleotide spans a precise signal joint. The AT2-5 oligonucleotide anneals 5 base pairs away on each side of a precise coding joint with two inosine residues in between to minimize possible effects of coding joints diversity on the intensity of the hybridization signal.
To determine the composition of signal joints, signal joints were PCR-amplified using 32 P-labeled Mark1 oligonucleotide, digested with ApaLI, and resolved by native PAGE, and the resulting bands were quantitated on a PhosphorImager. To determine the composition of coding joints, they were PCR amplified, agarose gel-purified, and cloned into pCR2.1 using a TA Cloning kit (Invitrogen). Clones were selected at random (and without prescreening by hybridization to AT2-5) for sequencing.

Cleavage in Trans in Vitro-
We have examined the ability of RAG1 and RAG2 proteins to cleave large plasmid-based substrates with 12-and 23-signals in cis and in trans in the presence of HMG2 in vitro. Placing the RSSs in trans may affect the reaction in two ways: first, the local concentration of one RSS in the vicinity of another RSS decreases; and second, the different topological location of the RSSs may affect the reaction via some unknown mechanism. To distinguish between these two possibilities, we constructed catenated substrates in which RSSs were located on two different circles that were not covalently attached but were topologically connected. In this way, the local concentrations of RSSs in the vicinity of each other remained high as in the case of the cis configuration, and only the topological effect, if any, could be investigated.
To create such a substrate, we introduced two ␥␦-resolvase sites (30) into a plasmid containing two RSSs to generate pAT2 ( Fig. 1). Control substrates, identical to pAT2 but lacking either the 12-RSS (pAT6) or the 23-RSS (pAT7) were also constructed. Treatment of supercoiled pAT2 with ␥␦-resolvase yielded two catenated circles (labeled A and B in Fig. 1), each containing one RSS, whereas pAT6 yielded the same products but with the 12-RSS missing from circle A. Although the ␥␦resolvase-mediated reaction did not go to completion, the RAGgenerated cleavage products from pAT2 and the catenated circles could easily be distinguished by Southern blots using probe AT2-2 and digestion with XcmI ( Fig. 1). Supercoiled pAT2 yielded a 6.8-kb band when linearized with XcmI (band 1 in Fig. 2), a 3.6-kb band when cleaved by the RAG proteins at only the 23-RSS (band 3), and a 3.1-kb band when cleaved at both RSSs or only at the 12-RSS (band 4). The catenated circle B (to which AT2-2 hybridizes) gives rise to a 4.2-kb band when linearized with XcmI (band 2 in Fig. 2) and a 0.9-kb band when cleaved at the 23-RSS (band 5).
In the standard reaction, substrates were treated with ␥␦resolvase and then incubated with RAG1, RAG2, and HMG2 in the presence of Mg 2ϩ . This was followed by deproteinization, digestion with XcmI, and analysis by Southern blotting with the AT2-2 probe. Controls included performing mock reactions with ␥␦-resolvase and/or the RAG proteins and performing the XcmI digest before cleavage with the RAG proteins (thus linearizing circle B and unlinking it from circle A).
In the absence of treatment with ␥␦-resolvase (Fig. 2, lanes [1][2][3][4][5][6][7][8][9], coupled cleavage at the two RSSs by the RAG proteins was indicated by the relatively prominent appearance of band 4 when both RSSs were present (lanes 4 and 7). A smaller amount of single cleavage at the 23-RSS was visible with either pAT2 or pAT6 (band 3, lanes 4, 5, 7, and 8), and no single cleavage at the 12-RSS was visible with pAT7 (lanes 6 and 9). As expected, in the absence of treatment with ␥␦-resolvase, digestion with XcmI before RAG-mediated cleavage had no effect on the pattern of cleavage (compare lanes 7-9 with lanes 4 -6). Treatment with ␥␦-resolvase resulted in conversion of more than half of the substrate molecules to the catenated form (compare bands 1 and 2 in lanes 10 and 11). RAG1 and RAG2 cleaved the 23-RSS on circle B more efficiently when circle A contained a 12-RSS than when it did not (band 5, compare lanes 12 and 13). The degree of the cleavage also depended on the local concentration of the complementary RSS: linearization of circle A with XcmI (lane 14) or of circle B with AflIII (data not shown) before RAG-mediated cleavage eliminated detectable cleavage at the 23-RSS. Efficient cleavage at the 12-RSS was also dependent on the presence of the 23-RSS on the catenated circle (data not shown).
We conclude that RSSs can undergo productive synapsis and coupled cleavage in trans. A similar conclusion was reached using oligonucleotide substrates at a concentration of 2 nM (15). It is clear that with plasmid DNA substrates at the concentration used in our experiments (0.45 nM), the two RSSs must either be in cis or constrained by catenation to undergo detectable synapsis.
V(D)J Recombination of Catenated Substrates in Vivo-To determine whether V(D)J recombination could occur in vivo between RSSs located in trans, we transfected cells expressing RAG1 and RAG2 with various plasmid substrates and measured the amounts of coding and signal joints formed by PCR. In most experiments, substrates were transfected by electroporation into the F2A1 cell line, which expresses truncated forms of RAG1 (amino acids 264 -1008) and RAG2 (amino acids 1-387). Cells were heat shocked 6 -10 h before electroporation to induce expression of RAG1 and RAG2, although omitting the heat shock produced equivalent results because of leaky expression of RAG1 and RAG2 in F2A1 (data not shown). Transfection efficiency was estimated by transfection with pCMV␤ and staining for ␤-galactosidase as described under "Experimental Procedures." Although substrates contained polyoma virus sequences necessary for replication, extrachromosomal DNA was harvested 14 -16 h post-transfection, before detectable replication occurs (33). The relatively early time of harvest also reduced the possible effects on detection of recombination caused by unequal stability of various recombination products.
Recovered DNA was quantitated by UV absorbance and by PCR with primers complementary to a nonrearranging portion of the plasmid (polyoma backbone). Results obtained by the two methods were in good agreement, and the amount of input DNA in PCR reactions was calculated based on its UV absorbance. However, we were not able to determine the proportion of recovered DNA that had actually entered the cell (substantial amounts of substrate adhere to the outside of the cells). We note that experiments were repeated multiple times, and good reproducibility in the results was observed between similar samples within a given transfection experiment. Furthermore, although variations in the efficiency of transfection or DNA  (12) were treated with various combinations of ␥␦ resolvase, RAG proteins, and XcmI, as indicated above the lanes, and samples were resolved on an agarose gel, transferred, and hybridized with probe AT2-2. Treated with ␥␦-resolvase, where included, preceded cleavage by RAG1 and RAG2. Cleavage with XcmI either preceded cleavage by RAGs but followed ␥␦-resolvase treatment (before), or followed cleavage by RAGs recovery will affect the absolute number of signal and coding joints obtained, they should not affect the ratio of coding to signal joints because both are measured in the same DNA preparation. Hence this CJ/SJ ratio can be compared between different samples and different transfections using only the assumption that the relative detection efficiency of coding and signal joints was the same for all samples for the given PCR primers. As negative controls, we mock transfected F2A1 (omitting the electric pulse) or transfected substrates into M12, the parental line of F2A1 that does not express RAG1 or RAG2.
In the first group of experiments, we transfected catenated substrates identical to those used in the in vitro cleavage experiments (that is, a mixture of catenated and uncatenated plasmids, because the ␥␦ resolvase reaction does not go to completion). We designed PCR primers (AT2a-f and Inne1 for coding joints, and AT2c-r and AT2b-r for signal joints; Fig. 1) that detected the recombination products of the catenated substrate but not those of the original noncatenated plasmid pAT2 (Fig. 1). PCR products were resolved on 5% native acrylamide gels, transferred to a membrane, and hybridized to AT2-5 or AT2-4 radiolabeled oligonucleotides specific for coding or signal joints, respectively. AT2-5 annealed to coding sequences on each side of and five base pairs away from the position of a precise coding joint and contained two inosine residues at the position spanning the joint, thereby reducing effects of coding joint diversity on the intensity of the hybridization signal. Blots were quantitated on a PhosphorImager.
We observed formation of both coding and signal joints using the catenated substrate with 12-and the 23-RSSs in trans (Fig.  3, lanes 3-6). Formation of these joints was dependent on the presence of RSSs on both catenated circles (data not shown). The catenated substrate was also digested with a restriction endonuclease that cleaved circle A, and in a separate reaction, with one that cleaved circle B, and the products of digestion were mixed together. This procedure yielded a sample containing the same circles as the catenated substrate but that were not catenated. Transfection of this sample also yielded coding and signal joints (Fig. 3, lanes 7-10). Together, the results demonstrate that coding and signal joints can be detected in trans and that this does not depend on catenation of the two recombining partners. We did not attempt to calculate a relative CJ/SJ ratio for these two samples. This experimental system was not explored further because it did not allow for a ready comparison of joining efficiencies in cis versus trans (different PCR primers were required for detection of joints with the two different configurations of RSSs).

Cis versus Trans and the Effect of Substrate Topological State on Recombination in Vivo-
To determine the relative efficiencies of coding and signal joint formation in cis and trans, we repeated the transfection assay described in the previous section but compared pAT2 (RSSs in cis) with a mixture of pAT6 and pAT7 (RSSs in trans). In this situation, the same pairs of PCR primers (Stop2 and Inne1 for coding joints and Mark1 and AT2c-r for signal joints; Fig. 1) could be used to amplify joints for both the cis and trans configurations. This strategy is essentially identical to that employed in a recent analysis of the same issue (26). We also explored the effect of substrate topology by transfecting supercoiled, relaxed circular, and linearized forms of the substrates. Relaxed circular and linear plasmids were generated by treatment of supercoiled plasmids with topoisomerase I and NcoI, respectively. The results of agarose gel electrophoresis of the various topological forms of the substrates are shown in Fig. 4.
The initial experiment revealed that relaxation of supercoiling had little if any effect on the relative efficiencies of coding and signal joint formation (Fig. 5A, compare lane 2 with lane 3 and lane 5 with lanes 6 and 7). In contrast, linearization led to a dramatic reduction in coding joint formation relative to signal joint formation (compare lane 2 with lane 4 and lane 5 with lanes 8 and 9), and linearization of both pAT6 and pAT7 had a greater effect in this regard than linearization of either one alone (compare lane 10 with lanes 8 and 9).
Placing the RSSs in trans resulted in a reduction in both signal and coding joint formation compared with the cis configuration ( Fig. 5A; note the different exposure times for coding

FIG. 3. Signal and coding joint formation in trans in vivo.
The substrate pAT2 was treated with ␥␦-resolvase to generate catenated circles and transfected into F2A1 or M12 cells as indicated above the lanes. In lane 1, DNA was mixed with F2A1 cells, but electroporation was omitted. In lanes 1-6, the catenated substrate was used. To generate the substrate for lanes 7-10, the catenated material was split into two portions, each of which was digested with a restriction enzyme that linearized one of the two circles, and the products of digestion were mixed. DNA was harvested from cells, quantitated, and subjected to PCR for coding joints (primers AT2a-f and Inne1) and signal joints (primers AT2b-r and AT2c-r) as described under "Experimental Procedures." PCR products were resolved on an agarose gel, transferred, and hybridized with oligonucleotide probes specific for signal and coding joints (see "Experimental Procedures"). The amount of DNA substrate added to each PCR reaction, in fmol of RSS, is indicated above the lanes, and the expected size and position of the PCR products are indicated at the left.  1, 4, and 7) were treated with either topoisomerase I or XcmI, as indicated above the lanes. The positions of the supercoiled plasmid, linear plasmid, fully relaxed circle (nicked), and partially relaxed closed circles (r.c.c.) are indicated at the right. joint detection in cis and trans). This was best visualized and quantitated in the experiment shown in Fig. 5B (compare lanes 2-5 with lanes [11][12][13], in which the amount of template added to the PCR reactions was varied, and a single exposure time for each assay is shown. PhosphorImager quantitation of these data revealed that placing the RSSs in trans reduced signal and coding joint formation 8-and 40-fold, respectively, compared with the cis configuration. However, as discussed above, it is more meaningful to compare the CJ/SJ ratio for the two RSS configurations, which for cis relative to trans yields a number of 5 (Fig. 5C). In other words, placing the two RSSs in trans results in a 5-fold reduction in coding joint formation relative to signal joint formation. Similar analyses revealed that linearization of the cis substrate (pAT2) resulted in a 20-fold decrease in the CJ/SJ ratio, whereas linearization of one or both of the trans substrates (pAT6 and pAT7) resulted in 25-or 200-fold decreases, respectively, in this ratio (Fig. 5C).
To determine whether these results were influenced by the use of truncated RAG proteins, heat shock induction of RAG expression or electroporation as a method of transfection, these experiments were repeated with the Dr3 cell line, which expresses heat shock-inducible full-length RAG1 and RAG2, and with the nonlymphoid CHO cell line. Truncated RAG1 and RAG2 were expressed in CHO cells from transient expression vectors that were transfected, together with the substrate plasmids, by lipofection (see "Experimental Procedures"). In all cases, we obtained results similar to those shown for the cell line F2A1 (data not shown).
We also ruled out the possibility that the results were biased in some way by the use of the AT2-5 oligonucleotide probe, whose region of homology spans the coding junction. When membranes were stripped and rehybridized with an oligonucleotide probe that hybridizes to the coding joint PCR product far (143 base pairs) from the coding junction, coding joint detection was identical to that with AT2-5 (data not shown).
Most (90%) signal joints formed in the experiments with F2A1 were precise fusions of the two RSSs, as determined by sensitivity of the PCR products to ApaLI digestion (data not shown). This was true regardless of substrate topology. Representative coding joints were cloned and sequenced for supercoiled and linear pAT2 (RSSs in cis) and for supercoiled pAT6 plus supercoiled pAT7 (RSSs in trans). In all cases, coding joints were found to have the expected structure, with deletion of small numbers of nucleotides from one or both coding ends (data not shown). We note that the clones sequenced were chosen at random and were not selected based on hybridization to the AT2-5 probe.
These results lead to two primary conclusions. First, synapsis, cleavage, and signal and coding joint formation can occur with RSSs located in trans. Second, substrate linearization (which does not alter the cis/trans configuration of the RSSs) has a much larger effect on the relative efficiency of coding joint formation than does changing from a cis to a trans configuration. A model to explain these results is presented below. DISCUSSION V(D)J recombination in trans is an important issue because of its potential involvement in a variety of chromosomal rearrangements leading to the development of neoplasias (19 -23, 34). However, it remains difficult to study V(D)J recombination in trans directly using chromosomal substrates, primarily because of the low efficiency of the reaction.
In this paper, we show that the cleavage step of V(D)J recombination can be performed in vitro in the presence of Mg 2ϩ and HMG2 with a substrate containing 12-and 23-RSSs in trans. Cleavage is stimulated by the presence of the appropri-

FIG. 5. Effect of cis versus trans position of RSSs and of the topological state of the substrate on recombination in vivo.
A and B, substrates pAT2 (cis) or pAT6 and pAT7 (trans), either in supercoiled, relaxed circular, or linear form, were transfected into M12 or F2A1 cells as indicated above the lanes. Samples were processed and analyzed as described in the legend to Fig. 3, except that PCR for coding joints was performed with primers Stop2 and Inne1 and for signal joints with primers Mark1 and AT2c-r. In A, 50 ng of recovered DNA was used in each PCR reaction, whereas the amount added to each reaction is indicated above the lanes in B. Note that in A, different exposure times were used for detection of coding joints in lanes 1-4 (1 h) and lanes 5-10 (22 h), whereas in B, the coding joint data displayed were obtained from a single exposure (exposure time 6 h). C, summary of the effects of cis, trans, and topological state of substrates on the coding joint to signal joints ratio. In the third column is displayed the CJ/SJ ratio for one substrate (X) divided by the CJ/SJ ratio for another substrate (Y), whereas the first two columns show the structure of X and Y for that row. Hence, for example, the first row shows the CJ/SJ ratio for the cis-circular substrate divided by that for the trans-circular substrates, indicating a 5-fold increase in coding joint formation relative to signal joint formation for the former compared with the latter. ate partner RSS in trans and, based on previous studies, is likely to occur in the context of a synaptic complex containing the two RSSs (15,35). Efficient cleavage in trans was also obtained using oligonucleotide substrates at RSS concentrations somewhat higher than those used in our experiments (2 nM versus 0.45 nM) (15). Taken together, the results suggest that the reduction of cleavage efficiency we observe for RSSs on unlinked molecules is due to the decrease in RSS local concentration and not to their topological (cis versus trans) location.
Given that cleavage in vitro can occur efficiently in trans, it was possible that V(D)J recombination in vivo might also occur in trans. An early study attempted to recover the dimeric circle that would result from recombination in trans between two extrachromosomal substrates but failed after screening 40,000 recovered plasmids (24). The authors of this study did not draw strong conclusions based on these results, but more recently, the results were cited as evidence against V(D)J recombination in trans (26). This does not appear to be a valid argument. The recombination frequency in the PD31 cell line used was sufficiently low that only 150 recombinants would have been expected for RSSs in cis, and several factors would have reduced that number for the trans configuration: (i) many cells would likely have taken up only one of the two plasmids because of the poor transfectability of this cell line; (ii) even in those cells that obtained both plasmids, our results and those of Han et al. (26) indicate that the events leading to signal joint formation would be reduced in efficiency by an order of magnitude because of the trans configuration; and iii) the only plasmids that would have been scored as trans-recombinants were those that contained both a signal and a coding joint, and prior evidence indicates that this can be quite inefficient because the efficiency of joint formation after cleavage is not 100% (36,37). Hence, it is likely that the methods employed in this early study were not sufficiently sensitive to detect trans recombination. V(D)J recombination between antigen receptor loci located on different chromosomes has been detected in normal lymphoid cells (reviewed in Ref. 1), but the low frequency of these events (on the order of one/10 5 lymphocytes) is consistent with the possibility that they occur subsequent to a chromosomal translocation that puts the two loci in cis (1). Recombination between the murine Ig and loci has been detected in an Abelson murine leukemia virus transformed pre-B cell line, at a frequency about 1000-fold lower than rearrangements in cis (25). These events were unusual in that they were mediated by two 12-RSSs and generated atypical, imprecise signal joints. Overall, these studies indicate that V(D)J recombination between RSSs located on different chromosomes is a rare event.
Our results demonstrate that extrachromosomal substrates containing RSSs in trans can successfully recombine in vivo to yield both signal and coding joints. This was true for RSSs on catenated, circular substrates and for RSSs on unlinked DNA molecules. We conclude that there is nothing intrinsic to the V(D)J recombination reaction that prevents synapsis, cleavage, and joint formation with RSSs located in trans. This, together with the very low frequency of interchromosomal V(D)J recombination events, suggests that regulatory mechanisms exist to suppress V(D)J recombination between different chromosomes. We infer that these mechanisms do not function, or function inefficiently, with transiently transfected extrachromosomal elements. Although the nature of these mechanisms are unknown, it is reasonable to think that they act to inhibit synapsis of inappropriate RSS pairs, either those on different chromosomes or in different antigen receptor loci (38).
How is it that, with extrachromosomal substrates, RSSs in trans undergo synapsis and cleavage at levels only moderately reduced (less than an order of magnitude) compared with the cis configuration? One possibility, raised by the fact that the 12-RSS and 23-RSS in pAT7 and pAT6 are contained in nearly identical plasmid backbones, is that pairing of homologous sequences between different plasmid molecules increases the effective local concentration of the two RSSs. This notion is consistent with the finding that the human homologous pairing protein HPP-1 is an important component of an in vitro end joining system that preferentially joins homologous ends (39). It will be important to determine whether the efficiency of trans V(D)J recombination is altered when the two RSSs reside in plasmids lacking homology.
Using circular substrates, placing the RSSs in trans resulted in roughly 8-and 40-fold decreases in signal and coding joint formation, respectively, compared with the cis configuration. The decrease in signal joint formation that we observe is in good agreement with the 5-10-fold decrease reported in a recent study by Han et al. (26). This study, which was quite similar to ours in design, also assessed the efficiency of RAGmediated cleavage of the transfected substrates by measuring the abundance of signal ends. Substrates with RSSs in trans yielded about 10-fold fewer signal ends than did a substrate with RSSs in cis, a decrease that mirrored the drop in signal joints (26). A simple explanation of these results is that RSSs in trans are less likely to synapse and undergo cleavage, resulting in the parallel drop in cleaved products and signal joints.
Surprisingly, the aforementioned study found that intermolecular coding joint formation was at least 1000-fold less efficient than intramolecular coding joint formation, yielding a greater than 100-fold drop in the CJ/SJ ratio (26). In contrast, we find only a modest 5-fold decrease in this ratio. Each study examined lymphoid and nonlymphoid cell lines and truncated and full-length RAG proteins. There are no obvious methodological differences between the two studies apart from the method used to recover transfected plasmid DNA (alkaline lysis followed by Qiagen spin column purification in this study; a nondenaturing Hirt method in the study of Han et al.), and this does not appear to provide a straightforward explanation for the differences in the results.
To explain the dramatic drop in intermolecular coding joints formation in trans, Han et al. postulated the existence of a mechanism capable of discerning whether two RSSs are in cis or in trans at an early step in the reaction and transmitting this information to the postcleavage complex in such a way as to allow signal but not coding joint formation. Because antigen receptor loci are typically quite large, such a mechanism would have to be able to distinguish between synapsis of two RSSs hundreds of kilobases apart on the same chromosome from synapsis of two RSSs on different chromosomes.
Our results lead to a quite different model in which there is no intrinsic difference between intermolecular and intramolecular recombination. Rather, we propose that the specific reduction in coding joint formation we observe when two RSSs are in trans or on linear substrates is explained by inefficient retention of coding ends, but not signal ends, in the postcleavage complex. This is consistent with biochemical experiments demonstrating that signal ends are bound much more tightly than coding ends after cleavage (14,18). Our model predicts frequent accidental release of a coding end from the postcleavage synaptic complex. In addition, it predicts that this coding end can be recaptured by the synaptic complex and successfully incorporated into a coding joint (Fig. 6). The effects we observe with a trans configuration of RSSs or with linearized substrates can be explained by a decreased efficiency of recapture of the released coding end with these substrates.
Consider first the 5-fold decrease in the CJ/SJ ratio observed for recombination in trans compared with cis. The recapture of a coding end is influenced by the local concentration of the end in the vicinity of the synaptic complex. In a simple model, this in turn depends on the length of the DNA segment connecting the coding end to the signal end, which is acting to tether the coding end to the synaptic complex. With circular substrates, the cis and trans configurations give rise to structurally very similar putative postcleavage complexes (Fig. 6, A and C), with one important difference: in the cis configuration, the lengths of the "tethers" are 6.4 and 0.4 kb, whereas for the trans configuration, the lengths are both 6.8 kb. Theoretical and empirical considerations (40) indicate that the difference between 0.4 and 6.8 kb for one end can account for the 5-fold decrease in the CJ/SJ ratio.
The model provides a straightforward explanation for large reductions in the CJ/SJ ratio caused by substrate linearization. The linearized cis substrate (Fig. 6B) gives rise to a putative postcleavage complex in which one of the coding ends is no longer tethered to the complex. If this coding end is released, it would be recaptured very inefficiently, if at all, resulting in a complex that could yield a signal joint but not a coding joint. Exactly the same situation arises if one of the trans substrates is linearized, and under both of these circumstances we observe a 20 -25-fold reduction in the CJ/SJ ratio (Fig. 5C). An even more severe defect is predicted if both of the trans substrates are linear, because in this situation neither coding end is tethered to the postcleavage complex (Fig. 6D). Indeed, we observe a 200-fold decrease in the CJ/SJ ratio in this instance (Fig. 5C).
It is possible that the loss of supercoiling in some way contributes to the reduction in coding joint formation observed with linear substrates. We attempted to address this possibility by transfecting relaxed circular substrates. Such substrates behave exactly like their supercoiled counterparts, suggesting that supercoiling is not a relevant factor in joint formation. However, we cannot rule out the possibility that the relaxed circles are rapidly supercoiled upon entry into the cell. Nor can we eliminate the possibility that the linear molecules are rapidly circularized inside the cell, although this seems unlikely because they yield results quite different from circular substrates. It is unlikely that the linear substrates are inherently unstable inside the cell (within the relatively short time frame of the experiments) because they support efficient signal joint formation.
An alternative explanation for the greatly reduced CJ/SJ ratio observed with linear substrates is that the presence of free ends on the substrate in some way destabilizes the postcleavage complex, thereby inducing coding end release. This would imply that frequent coding end release occurs only with linear substrates. A related possibility is that linearization could facilitate binding of DNA repair factors (e.g. a nuclease or components of the DNA-dependent protein kinase), and this could interfere specifically with coding joint formation by some unknown mechanism. These alternatives would not explain the 5-fold decrease in the CJ/SJ ratio observed in trans versus cis recombination.
An important implication of our model and the 20-fold reduction of the CJ/SJ ratio upon linearization of the cis substrate is that at least 95% of coding joints formed must have been generated after at least one coding end release-recapture event. This in turn implies that the efficiency of coding joint formation is significantly influenced by the overall stability of the postcleavage complex (which reflects the rates of coding end release and recapture). Recent experiments by Agard and Lewis provide strong support for this hypothesis. 2 They demonstrate that cryptic RSSs (thought to bind the RAG proteins poorly) specifically reduce the efficiency of coding joint formation in vivo and do so by acting at a postcleavage step of the reaction. Such cryptic RSSs are proposed to destabilize the postcleavage complex within which important processing reactions (such as coding end hairpin opening) occur. Combined with our findings, it would appear that reducing the time spent in the postcleavage complex (either by increasing the release or reducing the recapture of coding ends) leads to reduced coding joint, but not signal joint, formation.
During V(D)J recombination of endogenous antigen receptor genes, one or both of the coding ends is not tethered to a signal end, and yet coding joints appear to form efficiently and quickly. In particular, coding ends are difficult to detect in normal lymphoid precursors and in cell lines actively performing recombination (7,(41)(42)(43). Together with our results and the considerations noted above, these findings strongly suggest that mechanisms exist to stabilize the postcleavage complex during recombination of endogenous antigen receptor loci and that these mechanisms do not act or act inefficiently with 2 E. A. Agard and S. M. Lewis, submitted for publication. Signal ends are assumed to be held tightly in the complex. In contrast, coding ends are proposed to be readily released, with the efficiency of recapture depending on whether the coding end is tethered to a signal end, and if so, how long the tethering piece of DNA is. If one or both coding ends are released and not recaptured, only signal joint formation can occur. A, in cis with a circular substrate. B, in cis with a linearized substrate. C, in trans with circular substrates. D, in trans with linearized substrates. Note that the postcleavage complexes in A and C differ only in the lengths of the DNA strands linking the coding ends to the signal ends (0.4 and 6.4 kb in A and 6.8 and 6.8 kb in B). transiently transfected substrates. These mechanisms may function to prevent the release of coding ends or may act to retain released coding ends in the vicinity of the synaptic complex for rapid recapture. Intriguingly, deletion of the T-cell receptor ␤ enhancer from the mouse germline preferentially interferes with coding joint formation compared with signal joint formation, leading to the suggestion that factors associated with the enhancer may act to stabilize coding ends in the postcleavage complex (44).
Our ability to detect significant amounts of both coding and signal joints in trans further supports the possibility of a direct involvement of V(D)J recombination in the generation of chromosomal translocations. The rate of V(D)J recombination in trans on chromosomal substrates, and especially of the relevant loci, remains to be determined. We expect that it will be significantly lower than on extrachromosomal circles. The rate of loss and recapture of coding ends of chromosomal substrates by the synaptic complex also remains to be investigated.