|Home | About | Journals | Submit | Contact Us | Français|
Tn5 transposase cleaves the transposon end using a hairpin intermediate on the transposon end. This involves a flipped base that is stacked against a tryptophan residue in the protein. However, many other members of the cut-and-paste transposase family, including the RAG1 protein, produce a hairpin on the flanking DNA. We have investigated the reversed polarity of the reaction for RAG recombination. Although the RAG proteins appear to employ a base-flipping mechanism using aromatic residues, the putatively flipped base is not at the expected location and does not appear to stack against any of the said aromatic residues. We propose an alternative model in which a flipped base is accommodated in a nonspecific pocket or cleft within the recombinase. This is consistent with the location of the flipped base at position −1 in the coding flank, which can be occupied by purine or pyrimidine bases that would be difficult to stabilize using a single, highly specific, interaction. Finally, during this work we noticed that the putative base-flipping events on either side of the 12/23 recombination signal sequence paired complex are coupled to the nicking steps and serve to coordinate the double-strand breaks on either side of the complex.
Antibody and T-cell receptor (TCR) diversity is generated by V(D)J recombination initiated by the RAG proteins, RAG1 and RAG2. The recombination signal sequences (RSSs), where recombination takes place, have a distinctive arrangement resembling transposon ends. The relationship between V(D)J recombination and transposition was established beyond doubt by the discovery of RAG-mediated transposition and by the identification of a triad of conserved active-site residues. This evidence placed RAG1 firmly within the family of transposases and retroviral integrases that have a characteristic DDE triad of amino acid residues that coordinate catalytic metal ions in the active site (1, 26, 30, 35, 39, 46). Later, the Transib family of transposons was identified as the likely ancestral group of RAG1 (33).
In V(D)J recombination, the RAG proteins excise the DNA between a pair of RSSs. This fragment is the equivalent of an excised transposon, and it takes no further part in the canonical V(D)J recombination reaction. Instead, the variable regions of the genes encoding antibodies and TCR are created by the imprecise rejoining of the flanking DNA, referred to as the “coding flank.” A key feature of the cleaved coding flanks is that they have covalently closed hairpin ends. The asymmetric resolution of these hairpins contributes to the diversification of the coding sequences during rejoining. The hairpins themselves arise as a consequence of the molecular mechanism RAG-mediated RSS cleavage.
The crystal structure for the catalytic core of the human immunodeficiency virus type 1 integrase protein revealed a structural fold shared in common with RNase H and the Holliday junction resolving enzyme RuvC (22). RNase H and RuvC monomers each perform a simple nicking reaction that requires a single phosphoryl transfer event. Cut-and-paste transposition, which requires at least three phosphoryl transfer steps at each transposon end, therefore presents a mechanistic challenge. One solution to this challenge was revealed by the discovery of the DNA-hairpin cleavage-intermediate in V(D)J recombination and Tn10 transposition (Fig. (Fig.1)1) (34, 57). However, it is interesting to note that the existence of this intermediate was first suggested by Coen and colleagues on the basis of the genomic scars produced by excision of the hAT family transposon Tam3 in Antirrhinum majus (14).
All DDE family transposases, including RAG1, cut the DNA to expose the 3′-OH at the end of the element (or RSS). However, the fate of the opposite strand and the order of strand cleavage events vary within the group (reviewed in references13, 18, and 55). Some enzymes, such as the retroviral integrases and the bacteriophage Mu transposase, nick and integrate the 3′-OH directly without second-strand cleavage. The cut-and-paste transposons, which cleave both strands of DNA, can be divided into two groups. With some notable exceptions such as the piggyBac element, most prokaryotic family members cleave the bottom strand of the recombination site first, whereas most eukaryotic members cleave the top strand first (8, 10, 20, 41, 47, 48, 77). For those family members with a hairpin mechanism, the inverted polarity of the first step dictates the reversal of all subsequent steps (Fig. (Fig.1).1). In consequence, most eukaryotic members of the family can achieve transposition with one less phosphoryl transfer reaction than the prokaryotic members, which are obliged to resolve the hairpin intermediate. The eukaryotic members can simply release the hairpin ends or, as in the case of RAG, hand them on to host factors for further processing (40).
Insight into the hairpin mechanism was provided by a crystal structure for the Tn5 transpososome, in which the penultimate base on the second, nontransferred, strand was flipped from the helix and stacked against a tryptophan side chain in the protein (Fig. (Fig.1)1) (19). The flipped base seemed to provide the steric freedom that is presumed to be required for making and resolving the hairpin intermediate. Two groups searched for a residue in RAG1 that performs a function equivalent to the stacking tryptophan in the Tn5 transposase (27, 45). This work identified several candidate residues on the basis of their respective mechanistic defects and their rescue by modified DNA substrates.
Here we have further assessed the candidate stacking residues using biochemical techniques previously used to study the dynamics of base flipping in Tn5 and Tn10 transposition (6, 7). We have identified a distortion at position −1 of the V(D)J coding flank DNA. It is introduced after the first nick at the RSS and is therefore reminiscent of the flipped base at the end of Tn5. The distortion is perfectly correlated with the ability of wild-type and mutant RAG-RSS complexes to perform the hairpin step of the reaction. We conclude that this base is probably equivalent to the flipped base in Tn5. However, none of the candidate aromatic residues seems to fulfill the function of the putative stacking tryptophan residue. We therefore propose a model in which base flipping in RAG recombination is significantly different from that in Tn5 transposition.
Canonical V(D)J recombination occurs within a 12/23 RSS paired complex (24, 36, 60, 72, 73). This restriction is known as the 12/23 rule. More recently a further restriction, the so-called “beyond 12/23” (B12/23) rule has been proposed to explain the exclusion of direct Vβ-to-Jβ joining in the TCR β region, despite the presence of appropriately oriented pairs of 12 and 23 RSSs (4, 21, 31, 32).
Little is known of the mechanisms that enforce the 12/23 rule or coordinate cleavage on either side of the complex. However, during this work, we observed that the coding flank distortion was coupled on either side of a 12/23 RSS paired complex: the distortion of a nicked coding flank is suppressed by an unnicked partner. We present a model and discuss the biological significance of this conformational coupling and its relevance to the B12/23 rule.
The oligonucleotides DAR39, DAR40, DAR42, DG61, DG62, DG4, and DG10 were used in this work, as described in reference 46. The following additional oligonucleotides were used (with sequences given in parentheses): QME27 (5′-TAAGACGTCGACGCGT), QME28 (5′-[IdU]AAGACGTCGACGCGT), QME29 (5′-GGATCCGGTTTTTGTTCAGGGCTGTATCACTGTG), QME211 (5′-GGATCCGGTTTTTGTTCAGGGCTGTATCACTG[IdU]G) (25), WTTOP (5′-ACGCGTCGACGTCTTACACAGTGATACAGCCCTGAACAAAAACCGGATCC), TOPFLANK (5′-ACGCGTCGACGTCTTA), TOPARM (5′-CACAGTGATACAGCCCTGAACAAAAACCGGATCC), OLIGO-LIG (5′-ACGCGTCGACGTCTTACACAGTGATACAGCCCTGAACAAAAACC), DAR39AAA (5′-GATCTGGCCTGTCAAACACAGTGCTACAGACTGGAACAAAAACCCTGCAG), DAR42AAA (5′-GATCTGGCCTGTCAAA), DAR40TTT (5′-CTGCAGGGTTTTTGTTCCAGTCTGTAGCACTGTGTTTGACAGGCCAGAT), DAR40TTα (5′-CTGCAGGGTTTTTGTTCCAGTCTGTAGCACTGTG-TGACAGGCCAGATC), DAR40TTβ (5′-CTGCAGGGTTTTTGTTCCAGTCTGTAGCACTGTGT), DG62ABA−1 (5′-CTGCAGGGTTTTTGTACAGCCAGACAGTGGAGTACTACCACTGTG[aba]AAGAC AGGCCAGATC), and DG62ABA+2 (5′-CTGCAGGGTTTTTGTACAGCCAGACAGTGGAGTACTACCACTG[aba]GTAAGACAGGCCAGATC). The standard uncleaved 12 and 23 RSS substrates were made by annealing DAR39/40 and DG61/62, respectively. In the prenicked versions, DAR39 was replaced by DAR42 and DG10, while DG61 was replaced by DAR42 and DG4. For 3′-end labeling with Klenow enzyme, the standard oligonucleotides were truncated by one residue to provide a fill-in site. To change the standard coding flank from 5′-TAA to 5′-AAA, DAR40TTT was annealed to DAR39AAA or DAR42AAA. The 23 RSSs with abasic residues (shown by “aba” in brackets in the above sequences) were made by annealing DG62ABA−1 or DG62ABA+2 with DG61, DAR42, and DG4 as required. Thymine dimer-modified substrates were made by annealing DAR40TTα or DAR40TTβ with DAR39AAA for the uncleaved RSS or with DAR42AAA and DG10 for the prenicked RSS.
All oligonucleotides were polyacrylamide gel electrophoresis (PAGE) purified as single-stranded DNA before use. If required, they were 5′-end labeled with polynucleotide kinase. They were annealed by heating to 96°C for 1 min in 20 mM HEPES (pH 7.0)-100 mM potassium glutamate followed by slowly cooling to room temperature. Radiolabeled oligonucleotides (25 pmol in 50 μl) were annealed with a 50% excess of cold partners. Double-stranded oligonucleotides were PAGE purified before use.
Substrates for cross-linking were internally labeled with a single 32P atom immediately 5′ or 3′ of the IdU substitution (given in brackets in the sequences above) (25). QME27 and QME28 were 5′-end labeled using polynucleotide kinase and individually ligated to QME29 using OLIGO-RAG as a complementary template scaffold. The full-length ligated products, QME27-29 and QME28-29, were purified by denaturing PAGE and individually annealed to WTTOP to provide uncleaved IdU-substituted 12 RSS substrates. For the prenicked substrates, WTTOP was replaced with the TOPARM and TOPFLANK oligonucleotides.
Murine RAG1 and -2 were expressed as fusions with glutathione S-transferase (GST) and purified from CHO cell culture as described previously (45). RAG1 core (positions 384 to 1008) was either wild type or had point mutations as specified in the text. RAG2 was the wild-type core sequence (positions 1 to 387). HMGB1 was purified as described previously (44).
Each 40-μl reaction mixture contained 25 mM HEPES (pH 7.0), 50 mM K glutamate, 0.1 mg/ml bovine serum albumin, 8% glycerol, and 0.4 pmol radiolabeled DNA, plus 1 μl of RAG1 or -2 (~75 ng each, at 0.75 pmol and 1.1 pmol, respectively). Coupled 12/23 reaction mixtures contained 0.08 pmol of radiolabeled partner plus 3.2 pmol of the unlabeled partner. Coupled reaction mixtures in addition contained 100 ng (4 pmol) of HMGB1 protein. The following additions were made as indicated: 2 mM CaCl2, 4 mM MgCl2, and 4 mM EDTA. After 5 min at room temperature (~25°C), 2.5 μg of sheared salmon sperm DNA was added and the mixture was incubated for a further 10 min. To check the extent of complex formation, 10 μl was supplemented with 2 μl 50% glycerol and analyzed by electrophoretic mobility shift assay (EMSA) using a 4% acrylamide-bis (29:0.8) gel in 0.5× Tris-borate-EDTA buffer. Electrophoresis was for 90 min at 25 mA at 4°C. The remaining 30 μl of the reaction mixture was treated with KMnO4 for 30 s and analyzed on a DNA sequencing gel as described previously (7).
The cross-linking strategy was similar to that described by Eastman and colleagues (25). Complexes were assembled as described above. Reactions were placed as a drop on a sheet of Parafilm and were exposed to a Stratagene transilluminator (312-nm tubes) from above for 13 min at a distance of 4 cm. Reaction mixtures were recovered into tubes, heated to 68°C for 10 min to denature the RAG proteins, and adjusted to 10 mM MgCl2 and 1 mM CaCl2. DNase I (2 μg) and micrococcal nuclease (0.2 μg) were added, and the mixture was incubated for 1 h at 37°C. Reaction mixtures were mixed with Laemmli loading buffer supplemented with 1% sodium dodecyl sulfate (SDS), heated for 10 min at 95°C, and analyzed by SDS-PAGE (7.5% acrylamide) with a stacking gel containing 1% SDS. The gel was dried and recorded by autoradiography. Prestained markers (NEB) were used to locate cross-linked products.
V(D)J recombination in vivo is governed by the 12/23 rule, as originally suggested by Early and colleagues (23). However, several exceptions to this rule have been documented: for example, 23/23 recombination detected in a mouse strain with a noncanonical immunoglobulin H locus (38) and the apparent cleavage of isolated 12 RSS sequences (49, 63). The 12/23 rule also operates in vitro in reactions using purified components (24, 36, 60, 72, 73). However, under these artificial conditions it is probably less well enforced, and cleavage is readily detected in reactions containing an RSS of one type. One investigation into the requirements for cleavage using maltose-binding protein-RAG fusion proteins concluded that there is a >15-fold preference for the 12/23 reaction over a 12/12 reaction and that this preference is enforced about equally by preferential synapsis and by control of the chemical steps required for hairpin formation (66, 67). The particular reaction conditions also affect the sway of the 12/23 rule. Cleavage of the 12 RSS alone can be increased significantly by the method of RAG protein purification, by extended incubation times, and by using Mn2+ in place of Mg2+, which is the physiological divalent metal ion.
In this work, we use the GST-RAG fusion proteins which are probably less tightly governed by the 12/23 rule than the corresponding maltose-binding protein-RAG fusion proteins. Cleavage of the 12 RSS in the presence of Mg2+ is quite efficient, as previously noted (58). Nevertheless, our reactions conform to the 12/23 rule when both RSSs are present in the reaction mixture prior to addition of the RAG proteins. When coupled 12/23 reactions were required, we added the 23 RSS in threefold molar excess to reinforce the natural preference for the 12/23 synapsis (Materials and Methods).
In the Tn5 transpososome, the T+2 base on the second strand is rotated almost 180° out of the helix during the cleavage step of the reaction (Fig. (Fig.1).1). However, distortion of the helix can be detected even before the start of the reaction by a mild permanganate sensitivity of thymidine residues in the uncleaved complex. Following the first nick, permanganate sensitivity of these residues increases markedly when base flipping opens the DNA helix in preparation for the hairpin step (7).
To investigate whether RAG employs base flipping, we began by probing the permanganate sensitivity of the second strand (Fig. (Fig.2a).2a). Complexes were formed in the presence or absence of divalent metal ion, as indicated, and treated briefly with permanganate. There was a slight, but reproducible, RAG-dependent increase in the sensitivity of the thymidine at position +10 of the second strand (henceforth “T+10”). This indicates that the RAG proteins bind to the 12 RSS in the absence of metal ions, even though such complexes are barely detectable in an EMSA. With the uncleaved substrate (left panel), Ca2+ caused a very slight, but reproducible, increase in the sensitivity at T+2 and T−1.
When the experiment was performed with a prenicked substrate, the mild sensitivity of residues T+10 and T+2 to permanganate was similar to the uncleaved substrate (Fig. (Fig.2a,2a, right panel). However, the sensitivity of T−1 was greatly increased in the presence of Ca2+ which supports stable complex formation, but not catalysis. The sensitivity of T−1 indicates a significant distortion of the DNA following the first nick, as has been suggested by others on the basis of similar evidence (2, 12, 50, 59, 68). If the distortion at T−1 equates to V(D)J base flipping, then its location in the coding flank mirrors the reversed hairpin polarity of the hairpin compared to Tn5.
If the 12 RSS is incubated with Mg2+ instead of Ca2+, cleavage takes place as the prenicked substrate is converted to the hairpin product (Fig. (Fig.2a,2a, lane 9). Permanganate sensitivity at T−1 is lost, showing that this base has regained the protective stacking interactions with adjacent residues in the hairpin product. This may seem counterintuitive. However, within the loop region of a hairpin, which is generally between two and four residues long, depending on the sequence, the bases can be stacked and may also engage in various noncanonical base-pairing interactions (e.g., see references 11 and 70 and the references cited therein). We confirmed that the DAR40 coding flank hairpin is indeed resistant to permanganate by testing an oligonucleotide with an identical sequence (see Fig. S1 in the supplemental material).
It is possible that the sensitivity of the T−1 base in the nicked complex is due to a generalized melting of the flanking DNA, rather than a highly localized distortion such as base flipping. We therefore designed a new 12 RSS substrate with thymidine residues at the first three positions of the coding flank (Fig. (Fig.2b).2b). When the prenicked complex was probed with permanganate in the presence of Ca2+, the sensitivity of T−2 and T−3 was low, indicating that the deformation is confined to the T−1 position (lane 4).
We also probed the reactivity of the opposite strand, but no significant changes in permanganate sensitivity were detected in the uncleaved and the nicked complexes (not shown). However, the strong signal arising from the large amount of intact substrate that remains in a footprinting reaction made it impossible to detect the permanganate sensitivity of the bases immediately adjacent to the nick.
After the first nick in Tn5 transposition the flipped base can be photo-cross-linked to the stacking tryptophan residue (7). Since positions T+2 and T−1 of the RSS can also be photo-cross-linked to RAG1 (25), we therefore wished to test whether these contacts were likewise established by a conformational change following the first nick. RAG complexes were assembled using uncleaved and prenicked substrates with IdU substitutions at positions T+2 and T−1. Following UV exposure, cross-linking was analyzed by SDS-PAGE (Fig. (Fig.2c).2c). Cross-linking was specific to RAG1 and depended on the IdU substitution and exposure to UV. The cross-linking signal was undetectable, or barely detectable, with the uncleaved substrate. However, a clear signal was obtained with the prenicked substrate (lanes 6 and 10). It therefore appears that following the first nick there is a significant conformational change in the RAG complex that moves the IdU residue within range of a cross-linkable group such as an aromatic amino acid side chain. In the case of Tn5, the site of T+2 cross-linking is the stacking tryptophan residue (7).
The permanganate sensitivity and cross-linking assays suggested that the T−1 residue in the 12 RSS might be the equivalent of the flipped base in Tn5. To further test this hypothesis, we designed a modified 12 RSS substrates in which T−1 and T−2 residues were covalently linked by a cis-syn thymine dimer incorporated during oligonucleotide synthesis. We refer to this as the α-modification (Fig. (Fig.3a).3a). We reasoned that the rigidity of this molecular tether would prevent the rotation of one residue relative to the other. Furthermore, the bulk of the thymine dimer would prevent the insertion of T−1 into a close-fitting binding pocket in RAG1, analogous to that which accommodates the flipped base in Tn5. We also prepared a second substrate, the β-modification, in which residues T−2 and T−3 were linked.
The thymine dimer-modified 12 RSS substrates were tested in 12/23 cleavage reactions (Fig. (Fig.3b).3b). In this experiment, the modified 12 RSS is radioactively labeled and the reaction is supplemented with a threefold molar excess of unlabeled 23 RSS. The α-tether between T−1 and T−2 inhibited the nicking step severely (lane 5) but was much less detrimental to hairpin formation with a prenicked substrate (lane 11). The β-tether between T−2 and T−3 did not affect either step of the reaction significantly (lanes 6 and 12). One potential caveat to any interpretation of this experiment is that the thymine dimer modification causes a kink in the DNA (52). The α-tether between T−1 and T−2 may therefore be unfavorable to the nicking step for reasons unconnected to base flipping. Nevertheless, the insensitivity of the reaction to the β-tether supports the notion that steric freedom at T−1 is particularly important. Finally, the significant hairpinning activity with the prenicked α-tether suggests that the T−1 base need not enter a highly specific binding pocket in the protein during this step. It appears that RAG and Tn5 transposase are significantly different in this respect. However, this is not completely unexpected. The −1 position of the V(D)J coding flank can be occupied by any one of four bases, whereas the flipped base in Tn5 and Tn10 is always a thymidine and can therefore be accommodated in a specific binding pocket. If the coding flank T−1 is indeed flipped, as it appears to be from the evidence already presented, perhaps it is accommodated in a nonspecific cleft, analogous to that seen in the structures of other enzymes that are able to flip different types of bases (Discussion).
In the presence of the HMG cofactor, assembly of the mixed 12/23 RSS complex is strongly favored over either of the two unmixed complexes, particularly the 23/23 pair (described above and see references 24, 29, 36, 60, and 73). This bias allows opposite sides of the paired complex to be experimentally distinguished simply by labeling only one of the two partners.
Very little is known about the structure of the synapsis or the identity of the residues in the DNA or protein that mediate the 12/23 rule. We therefore wished to ask whether the conformational change at T−1 is coupled on either side of the RAG complex. RAG complexes were assembled in the presence of Ca2+ ions using a radioactively labeled prenicked 12 RSS (Fig. (Fig.4a).4a). As demonstrated above, residue T−1 was sensitive to permanganate oxidation (lane 4). The sensitivity was unaffected by the addition of HMG protein (lane 5). However, when an uncleaved 23 RSS partner was provided in the reaction mixture during assembly of the complexes, permanganate sensitivity at T−1 was lost (lane 6). In contrast, when a prenicked 23 RSS was provided, the permanganate sensitivity was unaltered (lane 7). The permanganate sensitivity of the nicked 12 RSS is thus dependent on the state of the partner in the 12/23 synapsis.
The reciprocal experiment was also performed by assembling RAG complexes with a radioactively labeled prenicked 23 RSS (Fig. (Fig.4b).4b). The permanganate sensitivity of T−1 was similar to that in the 12 RSS, except that it was dependent on the HMGB1 cofactor required for RAG binding to the 23 RSS (compare lanes 4, 11, and 12). The permanganate sensitivity of the 23 RSS was abolished when an uncleaved 12 RSS partner was provided (lane 13), but not when the partner was prenicked (lane 14). It therefore appears that the distortion at T−1 is coupled on either side of the complex and can take place only after both sides of the complex have been nicked. Since it seems likely that this conformational change is a prerequisite for the hairpin step, we will, for convenience, sometimes refer to it as an “activation” step.
We wondered how the RAG proteins sensed nicking of the RSSs and how the signal was transmitted to the other side of the complex. For example, we wanted to determine whether perhaps the movement of T−1 provided a direct cue for the activation step. Alternatively, the nick may be sensed at another site by a change in the curvature, stiffness, or tension in the DNA, for example.
We assembled RAG complexes in which a nicked 12 RSS was paired with various 23 RSS partners that were either normal or incorporated abasic residues at the T−1 and T+2 positions (Fig. (Fig.4c).4c). The abasic sites were incorporated during oligonucleotide synthesis using the dSpacer modification that provides a normal sugar-phosphate backbone with a missing base. As before, the permanganate sensitivity of the nicked 12 RSS was lost when an uncleaved 23 RSS partner was provided (compare lanes 3 and 4). This behavior was unchanged by the presence of abasic sites at positions −1 and +2 (lanes 5 and 6). This shows that the uncleaved status of the partner RSS is not sensed directly by the presence of either of these residues in the helix.
We next wished to ask whether either of these residues signaled the nicked status of a nicked partner RSS. When a prenicked 23 RSS partner was provided, the permanganate sensitivity of the 12 RSS was maintained, as expected (lane 7). This behavior was unchanged by the abasic modification at position −1 (lane 8). The T−1 base is therefore not directly involved in signaling the nicked status of the RSS. However, an abasic residue at position +2 abolished permanganate sensitivity and the prenicked 23 RSS behaved as if it was uncleaved (lane 9). This result shows that T+2 is critical for transduction of the nicking signal, but it is of course unclear whether this is a direct or indirect effect.
Two groups have searched for aromatic amino acid residues in RAG1 that might perform the same function as the tryptophan residue in Tn5 transposase that stacks the flipped T+2 base (27, 45). Both groups searched for mutants that were nicking proficient but hairpin defective. Grundy and colleagues proposed W893 and W965 as likely candidates (27). In their conclusion, however, they slightly favored W956 because it was rescued by an abasic substitution at T−1 and because the hairpin step with the W893 mutation was rescued by Mn2+ and certain coding flank sequences. Lu and colleagues tested every aromatic residue in RAG1 and identified two additional residues, Y935 and F971, as candidates (45). However, they assessed the mutations using slightly different criteria and concluded that W893 was the best candidate stacking residue. Lu et al. eliminated Y935 as a candidate by showing that it was the hydroxyl group rather than the aromatic ring that was important for hairpinning. They also eliminated the W956 residue on the grounds that the W956A mutation was primarily a nicking mutant. We investigated this point further and found that the behavior of the W956A RAG1 mutant is complex (see Fig. S3 in the supplemental material). The W956A mutation is almost completely defective for nicking and hairpin formation with the 12 RSS alone, irrespective of whether it is uncleaved or prenicked. However, it provides about 30% hairpin formation, compared to the wild type, in 12/23 reactions provided that both RSS are prenicked (see Fig. S3, lane 8, in the supplemental material). A similar effect was noted by Grundy et al. (27) using Mn2+ in the reaction.
The four RAG1 mutations discussed above were next tested for their effect on the permanganate sensitivity of the coding flank T−1 residue (Fig. (Fig.5a).5a). In our standard reaction with a prenicked 12 RSS, all four mutants, including W956A, were defective for the hairpin step, as indicated below the gel. There was a perfect correlation between the hairpin defect and the conformational change required for permanganate sensitivity at T−1. The alanine substitutions of all four aromatic residues were permanganate resistant (lanes 3, 4, 5, and 8). However, the serine and threonine substitutions of Y935, which provide a replacement hydroxyl group and remain hairpin proficient, were permanganate sensitive (lanes 11 and 12). In contrast, the hairpin-defective leucine and phenylalanine substitutions of Y935, which preserve the original residue's chain length and aromatic characters, respectively, were permanganate resistant (lanes 13 and 14). It seems that all of the hairpin-defective mutants have lost the ability to distort the DNA at the T−1 position. T−1 distortion is therefore probably a prerequisite for the hairpin step and may correspond to base flipping in Tn5 transposition.
We also tested two DDE mutants. Permanganate sensitivity was lost in the triple mutant, where the DDE residues had been changed to NNQ (Fig. (Fig.5A,5A, lane 18). In contrast, the single E962A mutant remained permanganate sensitive (lane 16). Crystal structures for DDE superfamily members suggest that the catalytic metal ions contribute significantly to positioning the substrate in the active site (51, 75). The loss of permanganate sensitivity in the triple mutant probably reflects the loss of both divalent metal ions from the active site, in contrast to the probable loss of just one metal ion in the case of the single mutant.
In Tn5 transposase, mutation of the stacking tryptophan residue abolishes photo-cross-linking of the transposase to the flipped base (7). We therefore tested whether the remaining candidate stacking-residues, W893, W956, and F971, were required for RAG1 cross-linking by IdU substitution at positions −1 and +2 in a prenicked 12 RSS (Fig. (Fig.5b).5b). Cross-linking was not abolished by the alanine substitutions at any of these positions. Instead, each mutation reduced cross-linking at each position by a similar amount. This is evidence against a direct stacking interaction between either of these bases and any of the amino acid residues. However, the residual cross-linking of the mutant RAG proteins on the prenicked 12 RSS suggests that they have achieved a conformational change, compared to the uncleaved situation (Fig. (Fig.2c,2c, lanes 5 and 9), but that it falls short of that required for the distortion of the T−1 residue detected by permanganate sensitivity.
The polarity of the phosphoryl transfer reactions in transposition and V(D)J recombination are defined by the asymmetric location of the recombinase recognition sequences on one side of the recombination site (Fig. (Fig.6).6). We will refer to the different hairpin reactions as proximal (Tn5) or distal (RAG and hAT) to the respective binding sites. The most parsimonious way to switch between the proximal and distal hairpin mechanisms is for the recombinase active site to translocate between opposite strands across the major or the minor groove (51, 75). In two dimensions, this movement round the helix can be illustrated as a 180° rotation of the catalytic domain about the z axis of the DNA (Fig. (Fig.6,6, top and middle elements). This reorientation would have to be accompanied by retuning the protein-DNA contacts to accommodate the reversed polarity of the DNA strand. The relative positions of the DDE residues would also change with respect to the various chemical groups on the DNA. This is most easily appreciated by noticing how the first D residue exchanges position with the E residue above the nucleophilic oxygen atom. However, changing the relative positions of catalytic residues may not be as problematic as it might seem at first sight (51, 75). In the Tn5 transpososome, at least, the metal ions appear to be coordinated symmetrically, suggesting that the active site should accommodate strands of either polarity with minimal tuning required. More generally, the two-metal-ion mechanism appears to be versatile, and distantly related members of the family appear to accommodate DNA strands of opposite polarity (51, 75).
The flipped base, which lacks the structural symmetry of the active-site metal ions, would place a more severe constraint on the reversal of the hairpin polarity. Rotation of the active site, as it moved from one strand to the other, would place the hairpin binding pocket on the 5′ side of the scissile phosphate, instead of the 3′ side (Fig. (Fig.6,6, middle element). It seems likely that this location would affect its functions. In the Tn5 transpososome, the flipped base is important for all four chemical steps of the reaction. It is particularly important for hairpin resolution and the integration steps, when it appears to hold the second strand in position (3, 7, 37). Of course, neither of these steps is performed by the RAG proteins, which are only concerned with the nick and hairpin steps of the reactions. Even so, the new position of the flipped base on the transferred strand would represent a significant mechanistic innovation for Transib and the hAT transposons, which must still perform integration.
The most significant similarity between the Tn5 and the RAG recombination reactions is a conformational change following the first nick. In Tn5, it is detected as the permanganate sensitivity of the second base 3′ to the scissile phosphate (T+2). In RAG recombination, the first base 3′ to the scissile phosphate is permanganate sensitive (T−1). In the Tn5 transpososome, the conformational change also promotes photo-cross-linking between T+2 and the transposase. Furthermore, the cross-linking is abolished by mutating the stacking tryptophan residue to alanine. However, although alanine substitution of the candidate aromatic residues in RAG1 abolished permanganate sensitivity, cross-linking at positions T+2 and T−1 of the RSS was reduced but not eliminated (Fig. (Fig.5).5). This suggests that although the aromatic residues are required for the conformational change, they do not directly contact either of the bases.
The W956 candidate stacking residue is located just before the residue of the RAG1 DDE catalytic triad. This suggests that it is more likely to be equivalent to the Tn5 transposase probe residue (W323) that occupies a similar position and helps force the flipped base from the helix. Indeed, this possibility was mentioned by Grundy et al. (6, 7, 27). None of the present results are inconsistent with this proposition.
In the Tn10 transpososome, the sensitivity of the flipped base at T+2 is comparable to the sensitivity of T−1 in the 12 RSS (6). However, in Tn5 the flipped base is protected by the stacking tryptophan and/or its location within a hydrophobic binding pocket (7). Its extrahelical conformation is revealed by mutating the tryptophan residue to alanine. It is therefore possible that the flipped base in V(D)J recombination is really at position T+2 of the RSS, but that its extrahelical conformation is masked by stacking on an aromatic amino acid side chain or sequestration within a hydrophobic pocket in the protein. However, none of the mutations in the candidate residues revealed permanganate sensitivity of this residue, as might be expected if this was the case (Fig. (Fig.5a).5a). We therefore conclude that none of the RAG1 candidate residues are likely to perform a function equivalent to the stacking tryptophan residue in Tn5 transposase.
We would like to suggest that it is the base at position −1 on the second strand of the RSS that is flipped from the helix (Fig. (Fig.6,6, bottom element). The positions of the flipped bases in Tn5 and the RSS, just 3′ to the respective scissile phosphates, are consistent with their functional equivalence. However, in this model the flipped bases do not interact with homologous regions of the respective recombinases (Fig. (Fig.6,6, compare top and bottom elements). The respective recombinases both make important interactions with T+2 on the respective second strands. However, the difference is the position of the flipped base, which, unlike the catalytic metal ions, is intrinsically asymmetric and cannot therefore interact with the same region of the protein after the polarity of the reaction chemistry has been reversed.
Base flipping is a common theme in nucleic acid metabolism. It has evolved several times, and several distinct mechanisms have been documented (56, 62, 64, 65). Even if the flipped base in Tn5 represents an ancestral condition for all cut-and-paste transposons, there is no reason to suppose that the stacking interactions would have been conserved when the polarity of the reaction chemistry was reversed. RAG may instead resemble the T4 pyrimidine dimer DNA glycosylase, which Stivers refers to as an “atypical” enzyme in which the flipped base is held in a featureless protein crevice (65). This would be consistent with the cross-linking data and the assessment of the candidate stacking residues. Indeed, the −1 position in the coding flank can be occupied by purine or pyrimidine bases that would be difficult to accommodate in a highly specific binding pocket. These ideas are also consistent with numerous examples of “good” and “bad” coding flank sequences and their rescue by mispaired bases (e.g., see references 17, 54, and 74).
Finally, in our view it seems plausible that the distal hairpin mechanism, exemplified by RAG and the hAT family of transposons, evolved independently from an RNase H-like ancestor with a single-strand nicking activity. Some support for this idea is provided by a comparison of the Tn5 and hermes (hAT) transposase crystal structures (19, 28). In Tn5 transposase, the RNase H structural fold is interrupted by the insertion of an extra subdomain that provides the stacking tryptophan residue and the hydrophobic pocket that stabilizes the flipped base. The RNase H fold of hermes transposase is also disrupted by the insertion of an additional subdomain. The interruption is at the same point as in Tn5 transposase, but the inserted domain is structurally unrelated. The common point of insertion could be a coincidence. However, it seems likely that the inserted subdomain in hermes transposase performs a function related to the hairpin reaction. The absence of homology between the inserted subdomains in the hermes and Tn5 transposases, in our view, reflects the independent origins of the proximal and distal hairpin reactions.
Cleavage at an unsynapsed Tn5 end is mechanistically excluded by the monomeric state of the unbound transposase and the trans architecture of the synaptic complex. RAG1 also cleaves the RSS bound in trans using a single active site, but its initial binding as a dimer from solution allows the possibility of uncoordinated cleavage events (69). Indeed, RAG1 can probably nick an unsynapsed 12 RSS, but the hairpin step depends on synapsis and prior nicking of the partner end (21, 36, 73, 76). Very little is known of the mechanism that coordinates the hairpin step on either side of the complex. Our results suggest that the hairpin reactions are coordinated by conformational coupling of base flipping on either side of the complex. Together with a large body of preexisting knowledge, this is summarized by the model presented in Fig. Fig.77.
The first step of the reaction is the interaction of an RSS with dimer of RAG1/2 (Fig. (Fig.7,7, complex A). A second unbound RSS is next recruited into a paired complex (complex B). This is followed by a nicking of one of the two partner RSSs (complex C). This causes a conformational change in one-half of the complex that signals the nicked status of the bound RSS. However, the uncleaved status of the partner RSS prevents progression to the base-flipping stage which is required for the hairpin reaction. In our experiments, complex C is equivalent to those mixed complexes in which the T−1 base was resistant to permanganate treatment (e.g., the N12/UC23 complex in Fig. Fig.4c,4c, lane 4). Nicking at the second RSS promotes a matching conformational change in the other half of the complex (complex D). Both halves of the paired complex are now signaling the nicked status of the respective RSSs. There follows a second conformational change, indicated as the “activation” step, which is accompanied by base flipping at position T−1 (complex E). This conformational change affects both sides of the complex, and is detected in our experiments by the permanganate sensitivity at T−1. The complex is now proficient for the hairpin step (transesterification) which completes cleavage of the RSSs (complex F). Initially, the coding flanks are retained in the RAG complex. Since transesterification maintains the energy of the phosphodiester bond, a final conformation change is required to prevent the reversal of the reaction (complex G).
In Tn5, transposition-coupled cleavage is imposed by the monomeric state of the unbound transposase and the trans architecture of the synaptic complex. However, it seems inflexible compared to the conformational coupling and two-step activation of the RAG complexes in V(D)J recombination. Flexibility is important for the 12/23 rule and in the choice of recombining partners. For example, if RAG binding to a favorable RSS was stabilized by the one-sided conformational change that accompanies the first nick, it would provide an extended window to scan for a second favorable RSS, minimizing the number of unproductive nicks created while searching for recombining partners.
The B12/23 rule describes another form of restriction that excludes direct Vβ-to-Jβ joining in the TCR β region (4, 21, 31, 32). Joining of these segments is inefficient despite the presence of appropriately oriented pairs of 12 and 23 RSS. It appears that this restriction is imposed equally by the spacer and the coding flank sequences, which together abolish nicking of some Jβ segments almost entirely (21). Nevertheless, some of these slow nicking segments have been found to perform synapsis very well. Schatz and colleagues propose that dissociation of a Vβ/Jβ synaptic complex is faster than nicking of the Jβ partner and that the nicked Vβ substrate released from this complex can go on and recombine with a 5′Dβ substrate (21). In our model, the arrested Vβ/Jβ synaptic complex would be represented by complex C. This illustrates the importance of conformational coupling in the RAG complex. Even though all of the protein and DNA components have been assembled together in a single complex, the first double-strand break at one RSS is suppressed by the unfavorable partner RSS. This may well provide for the integration of different signals and flexible responses in V(D)J recombination.
This work was funded by grants from The Wellcome Trust to R.C. and from the NIH to D.B.R. (AI-36420).
We thank Y. Bigot and F. X. Barre for allowing J.B. to perform in their labs experiments requested in review. R.C. thanks Patrick Swanson for helpful discussions during the preparation of the manuscript.
Published ahead of print on 31 August 2009.
†Supplemental material for this article may be found at http://mcb.asm.org/.