The crystal structure of the complex between the N‐terminal DNA‐binding domain of Tc3 transposase and an oligomer of transposon DNA has been determined. The specific DNA‐binding domain contains three α‐helices, of which two form a helix–turn–helix (HTH) motif. The recognition of transposon DNA by the transposase is mediated through base‐specific contacts and complementarity between protein and sequence‐dependent deformations of the DNA. The HTH motif makes four base‐specific contacts with the major groove, and the N‐terminus makes three base‐specific contacts with the minor groove. The DNA oligomer adopts a non‐linear B‐DNA conformation, made possible by a stretch of seven G:C base pairs at one end and a TATA sequence towards the other end. Extensive contacts (seven salt bridges and 16 hydrogen bonds) of the protein with the DNA backbone allow the protein to probe and recognize the sequence‐dependent DNA deformation. The DNA‐binding domain forms a dimer in the crystals. Each monomer binds a separate transposon end, implying that the dimer plays a role in synapsis, necessary for the simultaneous cleavage of both transposon termini.
Tc3 of Caenorhabditis elegans is a member of the Tc1/mariner family of transposable elements. Members of that family are found in a wide variety of organisms, ranging from fungi to humans (Doak et al., 1994). Transposable elements are small stretches of DNA that can move from one position in the genome to another. The proteins responsible for the excision and insertion of the transposon into the genome are generally encoded by the transposon sequences. In the Tc1/mariner family, the transposons encode only a single protein, the transposase, that is capable of performing the entire transposition reaction in vitro (Lampe et al., 1996; Vos et al., 1996). The Tc1/mariner transposase genes are flanked by terminal inverted repeats. The sequences of the terminal inverted repeats are not conserved between different elements, apart from the four most terminal nucleotides. Another shared property of this family is that the transposon DNA is inserted into TA sequences of the host genome (for a review, see Plasterk, 1996).
The first step in transposition is the recognition of the transposon DNA by a specific DNA‐binding domain of the transposase protein. For Tc3A, the Tc3 transposase, it has been shown that the N‐terminal domain is responsible for this specific DNA binding. This domain binds a region ∼12 bases away from the DNA cleavage site (van Luenen et al., 1993; Colloms et al., 1994) (Figure 1). Between the N‐terminal specific DNA‐binding domain and the catalytic domain, another DNA recognition domain is present. This second, more C‐terminal domain recognizes DNA sequences located more towards the cleavage site (R.F.Ketting and R.H.A.Plasterk, unpublished results). Such a bipartite DNA binding has also been shown for the related Tc1 transposase (Tc1A) of C.elegans (Vos and Plasterk, 1994). In general, there is little sequence conservation in the N‐terminal regions of Tc1/mariner transposases and no apparent conservation of the inverted repeat sequence. Thus, the specific DNA‐binding domain of each transposase of this family recognizes the termini of its own transposon specifically, and the transposase protein of a given element will act only upon its own transposon ends.
Protein sequence alignments have revealed a weak similarity between the N‐terminal region of the Tc1‐like Minos element and a DNA‐binding domain of the Pax/paired family, the paired domain, found in mammalian and Drosophila genes (Franz et al., 1994). The paired domain is a conserved DNA‐binding domain found in a set of transcription factors (Pax proteins) that play important roles in development. The significance of the similarity is enhanced by the identification of the bipartite nature of the paired DNA‐binding domain (Czerny et al., 1993; Xu et al., 1995). The difference from the Tc3A bipartite DNA binding is that the C‐terminal part of the paired domain could not be shown to bind to DNA (Xu et al., 1995). However, in other members of the paired domain family (Pax proteins), a C‐terminal part plays a role in site‐specific recognition (Czerny et al., 1993; Epstein et al., 1994a,b; Xu et al., 1995). Furthermore, secondary structure elements are similar for the Tc1A and Tc3A DNA‐binding domains (as predicted with PHDsec, Rost and Sander, 1993).
Part of the catalytic domain is also involved in DNA binding. In a South‐Western assay, it has been shown that a region between amino acids 98 and 159 shows non‐specific DNA‐binding activity (Colloms et al., 1994). The catalytic domain of Tc3A shares sequence homology with the IS630‐Tc1 family and contains a catalytic triad, DDE motif, of carboxyl groups (Figure 1) (Doak et al., 1994), that is important for the Tc3 transposition activity (van Luenen et al., 1994). Other polynucleotidyl transferases also use three (sometimes four) carboxyl groups to position one or two catalytic divalent metal ions (Yang and Steitz, 1995). This family includes ribonuclease H (Doolittle et al., 1989), retroviral/retrotransposases (Fayet et al., 1990; Kulkosky et al., 1992), Mu transposase (Baker and Luo, 1994) and RuvC resolvase (Ariyoshi et al., 1994). The crystal structures of the catalytic domains of several of these proteins (Katayanagi et al., 1990, 1992; Yang et al., 1990; Ariyoshi et al., 1994; Dyda et al., 1994; Rice and Mizuuchi, 1995) revealed that they share a structurally related core in which the catalytic carboxyl groups are oriented in a similar way. It is likely that the catalytic domain of the Tc3 transposase has a similar three‐dimensional fold.
The detailed mechanism of transposition has many variations, for example in invoking single or double strand breaks of the host DNA, in the target site specificity or in the number of proteins involved (for a review, see Plasterk, 1995). Synapsis (assembly of multiple proteins, the transposon DNA ends and the target DNA) is thought to be required for proper transposition. This has been studied extensively for phage Mu, in which tetramerization of the transposase on the transposon ends is essential for transposition (for a review, see Chaconas et al., 1996). Another example of the importance of synapsis has been shown for the Moloney murine leukemia virus. When one of the two ends of this viral DNA is mutated to block cleavage, cleavage is inhibited at the other (wild‐type) end as well. This indicates that cleavage at either end (the first step in the integration process) only takes place when both ends contribute to the synapsis and are recognized by the integrase protein (Murphy and Goff, 1992).
To understand how the Tc3 transposase recognizes its own transposon DNA ends, we studied the structure of the N‐terminal specific DNA‐binding domain (Tc3A‐N) in complex with transposon DNA. This is the first crystal structure of a DNA‐binding region of a transposase in complex with DNA. With this, we can begin to understand the mode of recognition by transposases of their DNA substrates, and eventually the mode of binding and regulation of simultaneous cleavage at the two ends of the transposon.
Results and discussion
The overall structure of the protein–DNA complex
We have determined the X‐ray structure of the 65 amino acid DNA‐binding domain of Tc3 transposase (Tc3A‐N) in complex with transposon DNA (Figure 2A) at 2.45 Å resolution. The parts of the transposase and transposon used in co‐crystallization are indicated by shaded boxes in Figure 1. The structure shows that Tc3A‐N contains three α‐helices (residues 9–20, 25–32 and 36–44) typical of proteins sharing the helix–turn–helix (HTH) motif, such as homeodomains. The first helix is involved in dimerization of the protein domains in the crystal. Each monomer binds a DNA molecule. The second and third helix form the HTH motif and are involved in DNA recognition in the major groove. The N‐terminus of the protein interacts with DNA in the locally narrowed minor groove. The C‐terminus (residues 53–65 and the 6‐histidine tag) is not visible in the final electron density, probably due to flexibility. The 20/21 DNA oligomer is in a non‐linear B‐DNA conformation and both the major and minor groove interact with the protein.
The HTH motif makes base‐specific contacts in the major groove
Part of the recognition process of Tc3A for its own transposon DNA is mediated by direct readout of four bases by the HTH motif. The HTH motif, comprising helices 2 and 3, interacts in the major groove of the DNA. There are also extensive contacts with mainly one DNA backbone strand: 10 H‐bonds and seven salt bridges (Figure 2B and C).
Within the HTH motif, the N‐terminus of helix 2 is involved in extensive contacts with the phosphate groups in one DNA backbone strand. However, one side chain (His26) is making a purine‐specific hydrogen bond to G7 (for base numbering see Figure 1). Backbone amides of Leu25 and His26 make hydrogen bonds to phosphate group 7. The dipole moment (Hol et al., 1978) of helix 2 interacts favorably with phosphate group 7 as well. The contacts with one of the DNA backbones are consolidated by Ser24 (in the loop connecting helices 1 and 2) and Arg30, making hydrogen bonds to phosphate group 6. More salt bridges are made by His26 with phosphates 6 and 7 and by Arg30 with phosphates 5 and 6.
Two amino acids in the turn of the HTH motif make contacts with the other DNA backbone: Arg34 is hydrogen bonded to sugar 109 and salt bridged to phosphate 110. Ser35 donates a hydrogen to phosphate 110.
The N‐terminal part of helix 3 (the recognition helix in the HTH motif) is involved in base‐specific contacts with three bases (including one mediated via a water molecule) and several DNA backbone contacts. Arg36 makes one guanine‐ and one purine‐specific hydrogen bond to the guanine base at position 8. Arg40 makes a thymine‐specific hydrogen bond, via a water molecule, to T9. His37 recognizes G110 via a hydrogen bond to the purine‐specific N7 or to the guanine‐specific O6 of G110. We are not able to distinguish this at the current resolution. Helix 3 makes additional hydrogen bonds through Arg36 via a water molecule to phosphates 7 and 8 and through Cys38 to phosphate 110. Salt bridges are made by both Arg36 and Arg40 to phosphate group 8.
Minor groove interactions
Additional base‐specific recognition (of three bases closer to the DNA cleavage site) occurs by binding of the N‐terminus in the minor groove. Six more hydrogen bonds are made with the DNA backbone by both the N‐ and C‐termini of the domain (Figure 2).
The side chains of residues Pro2 and Arg3 are inserted in the minor groove. Since Met1 is missing (as shown by sequencing of the purified protein), Pro2 is at the N‐terminus and thus positively charged. The two positively charged N‐terminal residues bind like a thumb and finger between the negatively charged phosphate backbones of the locally relatively narrow minor groove (Figure 2D). Pro2 makes a pyrimidine‐specific hydrogen bond to T14 and a hydrogen bond to sugar 15. The rest of Pro2 is in van der Waals contact with sugars 15 and 108. Arg3 makes one pyrimidine‐specific hydrogen bond to T12 and one to sugar 12. The backbone amide of Arg3 most likely makes a purine‐specific H‐bond to A109, but a H‐bond to sugar 109 cannot be excluded. The backbone amide of Gly4 is hydrogen bonded to sugar 109 and the backbone amide of Ala6 to phosphate 16.
The last ordered amino acids of the C‐terminus (49–52) are close to the minor groove as well. Tyr49 OH and Ser52 Oγ are donating hydrogens to phosphate groups 109 and 108 respectively. There is some electron density visible beyond Ser52 in the minor groove, although it is not continuous and cannot be interpreted, indicating that the flexible C‐terminus is continuing along the minor groove towards the DNA cleavage site.
The Tc3A domain behaves like a 25 kDa protein on a gel filtration column, while the mol. wt is only 8.3 kDa. The 19 flexible C‐terminal amino acids are likely to cause a shift to higher molecular weight; however, the difference between the expected 8.3 kDa for a monomeric domain and the observed 25 kDa peak cannot be attributed to that only. It is more likely that Tc3A‐N is forming a dimer in solution.
In the crystal, we observe protein dimers as well, in which each protein monomer binds one DNA molecule. Helix 1 is involved in the dimerization of two protein domains by a 2‐fold crystallographic rotation. Each protein domain makes, at one side, contact with DNA and, at the opposite side, contact with a protein of a symmetry‐related complex (Figure 3). The accessible surface of a monomer of the Tc3A domain is 3999 Å2 (calculated with GRASP, Nicholls et al., 1991). Upon dimerization with another Tc3A domain, 12% (475 Å2) of this surface is buried on each monomer. The contact, which involves mainly helix 1 of each domain, is predominantly hydrophobic (24 van der Waals contacts), and only two hydrogen bonds occur (between the carbonyl oxygen of Ala13 of both molecules and the Nϵ of Gln14 of the symmetry‐related molecules). The dimer found in the crystals, and presumably in solution, may also be present in active transposition complexes.
DNA conformation and recognition through sequence‐dependent conformation
The DNA in the crystallized complex is not in a linear conformation (Figure 2A), but bent at both ends in different directions and planes. The bending of the DNA is stabilized (Strauss and Maher, 1994) by the large amount of positive charge on the protein, which is located mainly at the interface with DNA (Figure 4).
The conformation of the DNA was analyzed with the program CURVES (Lavery and Sklenar, 1989) using the global parameters. The average helical twist of 33.5° (10.7 residues per turn) and the average rise per base pair of 3.41 Å are typical for B‐DNA. The bends of the DNA are reflected in increases in the roll and tilt angles of the base pairs and in deviations of the major and minor groove widths and depths, compared with average B‐DNA (Stofer and Lavery, 1994). To define the sugar puckers, higher resolution data will be needed.
DNA sequences highly enriched in G:C base pairs have a tendency to form low twist angles (underwinding) and positive roll angles of the base pairs, resulting in a widening of the minor groove and narrowing of the major groove (Travers, 1993). The seven consecutive G:C base pairs (2:119–8:113) in this structure indeed have a relatively low twist angle (30.5° on average) and a positive roll angle (∼12° per base pair). The roll angles of this stretch add up to an 82° bend of the DNA. In the crystal, one end of each DNA molecule bends into the minor groove of a symmetry‐related DNA molecule, with the base pairs of the two DNA molecules almost perpendicular (Figure 5A). Part of this bending is visible in the top part of Figure 2A as a DNA bend to the right, but it is also partly away from the viewer.
At the position of the bent G:C stretch, Tc3A‐N makes numerous hydrogen bonds and salt bridges to only one of the DNA backbones (phosphates 5–8, Figure 2B and C). These contacts would not have been possible if the DNA was more linear. It is not clear whether the bend is present in the DNA alone, to which the protein adapts itself, or whether the bend is induced by the interaction with the protein. A third theoretical possibility, that the bend is caused or stabilized by the crystal contacts, seems unlikely given the exquisite complementarity of the protein and the deformed DNA.
The bending at the other end of the DNA oligomer (more towards the DNA cleavage site in the lower part of Figure 2A) is reflected in mainly negative roll and tilt angles (around base pairs 13:108 and 14:107) and a narrowing of the minor groove (4 Å compared with 6 Å for average B‐DNA; Stofer and Lavery, 1994). The bending is ∼30° and the direction is to the right in Figure 2A, and partly towards the viewer. This part of the DNA sequence contains AT bases, which have a tendency to form a narrow minor groove (Travers, 1993). In this narrow minor groove, the N‐terminus is inserted, making hydrogen bonds and van der Waals contacts with both DNA backbones (Figure 2D). Such intimate protein–DNA contacts would not be possible with a wider minor groove. Again, the Tc3A domain is complementary to the sequence‐dependent DNA conformation, contributing to the recognition by Tc3A of its own transposon DNA.
Unusual DNA–DNA crystal contacts
The DNA helix axis is roughly in the crystallographic b‐direction. This is visible in the diffraction pattern as a fiber diffraction pattern at 3.4 Å along the b‐axis. Although the DNA is mainly parallel to the b‐axis, its non‐linearity causes unusual crystal contacts.
The A1 and T120 bases at one end of the oligomer do not form the expected base pair, but are flipped outwards. These bases, together with the last base pair (G2–C119), are interacting in the minor groove and are almost perpendicular to the base pairs of a symmetry‐related DNA molecule (Figure 5A). Fraying of a DNA molecule at one end has been observed before (Joshua‐Tor et al., 1992), where the flipped‐out bases are also interacting in the minor groove of a symmetry‐related molecule. However, the base pairs of the symmetry‐related molecule are not perpendicular, as in this case.
The other end of the DNA forms a shifted semi‐continuous helix with the same end of a symmetry‐related DNA molecule (Figure 5B). These two molecules are related by a 2‐fold rotation axis in the a‐direction (perpendicular to the paper in Figure 5B). The overhanging base T21 makes a triple helix by forming a Hoogsteen base pair with the symmetry‐related base A101 in base pair A101:T20.
Comparison with other proteins
The three‐dimensional fold of the Tc3A domain is a HTH fold. Proteins sharing this motif are the eukaryotic homeodomains and transcription factors, prokaryotic repressors and Hin recombinase, amongst others (for a review, see Luisi, 1995; Wintjens and Rooman, 1996). The Tc3A domain is most similar to the N‐terminal part of the paired domain (Figure 6, Xu et al., 1995). The r.m.s. difference of 51 Cαs is 2.3 Å, according to the 3D alignment of the DALI‐server) http://www.embl-heidelberg.de/d…; Holm and Sander, 1993), and the r.m.s. difference of the Cαs in the three helices and connecting loops is only 0.66 Å. Other similar structures are, for example, the Oct‐1 POU domain (Klemm et al., 1994), C‐myb DNA‐binding domain (Ogata et al., 1995), engrailed homeodomain (Kissinger et al., 1990) and the paired homeodomain (Wilson et al., 1995). The three helices and connecting loops have r.m.s. Cα differences with Tc3A‐N of 1.3–1.5 Å.
Surprisingly, the recently determined Zn‐containing N‐terminal region of the functionally similar HIV integrase also contains a comparable HTH motif (Cai et al., 1997; Eijkelenboom et al., 1997). Superposition with the three helices and connecting loops of Tc3A‐N results in a Cα r.m.s. deviaton of 2.0 Å. The C‐terminus, however, folds in a different direction, where it provides two of the ligands to the Zn ion, which is coordinated further by a histidine in helix 1 and a histidine in the loop between helices 1 and 2. Data on the function of this domain are not clear, and it is unknown whether it is involved in DNA binding, although the presence of the HTH domain is suggestive.
The main differences between the N‐terminal part of the paired domain and Tc3A‐N are located in their N‐ and C‐termini. The few residues preceding the first helix in the Tc3A domain adopt a conformation different from that of the longer N‐terminus of the paired domain, which forms a small β‐sheet (Figure 6). Helix 3 is longer in the paired domain than in the Tc3A domain, but the direction of the C‐termini is similar. In the paired domain, the C‐terminus is connected to the second domain, but the linker is not clearly visible in the electron density (Xu et al., 1995). In our crystals, only the loop towards the second DNA recognition domain (residues 53–65) is present, and it is invisible in the electron density as well.
Docking on DNA
It has been described earlier that there are several ways of docking an HTH motif on DNA (for example, see Suzuki and Gerstein, 1995; Wintjens and Rooman, 1996). The Tc3A‐N protein is very similar to the N‐terminal part of the paired domain and homeodomains. These have, however, a different way of docking on DNA. The homeodomains have only helix 3 (the recognition helix) inserted in the major groove, and the residues in the center of this helix are interacting with DNA. This recognition helix is relatively long and there are common features in the amino acid sequence playing a role in DNA interaction (Suzuki, 1993). The paired domain belongs to another family, which also includes the prokaryotic Hin recombinase and λ repressor. In this family, both helices 2 and 3 interact with DNA, and this family is distinguished by a relative short helix 2 of which the N‐terminus interacts with the DNA backbone. Helix 3 is also interacting in the major groove, but at a different angle compared with the homeodomain family (Kissinger et al., 1990; Xu et al., 1995). The HTH motif of the Tc3A domain docks in a variant of the paired/Hin/λ family, although the non‐linear DNA in the Tc3A complex is exceptional.
There are nine residues (18%) in Tc3A‐N that are identical to the N‐terminal part of the paired domain. None of these residues are present at equivalent positions in the Hin recombinase (Feng et al., 1994) or the λ repressor (Beamer and Pabo, 1992). Only three of these residues (Arg30, Ser35 and Cys38 of Tc3A) are involved in similar DNA contacts. None of these are sequence specific; all three interact with the phosphate groups in the DNA backbone. The absence of conserved residues that interact with DNA is in contrast to the homeodomain family, which has anchor residues.
The locations within the HTH fold of the side chains that interact with the bases of the DNA are not very conserved either. Some side chains are located at the N‐terminus of helix 3, but not at identical positions. Others are situated towards in the center of helix 3, but again not at a fixed position. Tc3A‐N has one exceptional side chain in helix 2 (His26) that is involved in a base contact.
Correlation with biochemical data
The double‐stranded DNA oligomer used in this study was based on methylation interference (12 bp from G7 to A18) and footprinting (∼20 bases on each strand) studies (Colloms et al., 1994). The adenine and guanine bases which showed strong methylation interference are indeed in contact with the protein. A few bases with weak interference (G16, A17 and A18) are located where the flexible C‐terminus of Tc3A‐N is pointing. The middle of the 20/21 oligomer in the crystal is in contact with the protein, both in the major and minor groove, consistent with the prediction based on the methylation and footprinting studies (Colloms et al., 1994).
The same 65 amino acid N‐terminal domain with a histidine tag at the N‐terminus (instead of the C‐terminus as used in this study) did not bind transposon DNA, as shown by gel retardation assays (data not shown). The N‐terminal residues of the protein (Pro2 and Arg3) are bound in the minor groove, and a histidine tag at the N‐terminus would probably interfere with these contacts, making a protein–DNA complex impossible. Whether the N‐terminal Met1 (absent in this study) is also absent in vivo in C.elegans is unclear. The base‐specific contacts of Pro2 could be disturbed if Met1 were present, suggesting that Met1 is also absent in vivo.
We were not able to see the amino acids beyond Ser52, although some electron density in the minor groove suggests that the chain is continuing along the minor groove in analogy with the loop in the paired domain. A smaller N‐terminal part of Tc3A (amino acids 1–54) has been shown, however, not to bind to Tc3 transposon DNA (Colloms et al., 1994). This indicates that at least part of the floppy end (54–65) is essential for proper folding of this domain and/or DNA binding.
Comparison with Tc1 and Tc1A
Although the N‐terminal DNA‐binding domains of the related Tc3A and Tc1A show no obvious homology, secondary structure prediction (PHDsec, Rost and Sander, 1993) of the N‐terminal domain of Tc1A predicts three α‐helices (residues 12–23, 28–35 and 39–49), extremely similar to the Tc3A domain, strongly suggesting a HTH motif. Ten amino acids (out of the 65 of Tc3A) are identical in both proteins, when they are aligned on their (predicted) secondary structures. Most of these residues do not interact with DNA, except Arg34 and Ser35 in Tc3A (Arg37 and Ser38 in Tc1A), making DNA backbone contacts. The DNA backbone contact of Ser35 in Tc3A was also observed at the equivalent position in the paired domain.
DNase footprints and methyl interference studies have shown that Tc1A and Tc3A bind to the transposon ends at (almost) identical distances from the DNA cleavage site (Vos et al., 1993; Colloms et al., 1994). However, the DNA sequence that is recognized by the Tc1A DNA‐binding domain is not at all similar to the one recognized by Tc3A. Only three base pairs in the Tc1 and Tc3 transposon ends that are contacted by the N‐terminal domains are identical, when aligned from the DNA cleavage site. These base pairs (T12:A109, A13:T108 and T14:A107 in our numbering system) are in the region of the narrow minor groove where the N‐terminus of Tc3A is interacting. However, Tc1A has a three amino acid longer N‐terminus when it is aligned on (predicted) secondary structure and, therefore, probably does not have an analogous mode of recognition.
Prediction of other transposase DNA recognition modes. The crystal structure shows that recognition of Tc3 transposon DNA by the Tc3 transposase is based on a combination of sequence‐specific hydrogen bonds as well as a remarkable complementarity between deformed DNA and the protein. In the complex, the transposon sequence has a non‐linear character with irregular major and minor grooves. This DNA conformation is stabilized by extensive interactions with the protein, which could not be present if the DNA was linear. Thus, the protein either probes the pre‐set deformation of the DNA, or possibly makes use of the ability of this particular DNA sequence to change its conformation as a response to this interaction. Further specificity is given by a number of base‐specific hydrogen bonds, in both the major and the minor groove of the DNA.
As predicted, the Tc3A‐N protein has the same fold as that found for the N‐terminal part of the paired domain in the paired/Pax family of proteins. The two proteins also dock on DNA in the same location and orientation. However, very few specific contacts have been conserved, and the impressive complementarity to sequence‐dependent deformation of Tc3 was not seen in the paired domain, where the DNA is almost linear. This limits the possibilities of predicting DNA binding for other transposases. For the Tc1/mariner family, the weak sequence homology and the similar secondary structure prediction results lead to the expectation that their N‐terminal DNA‐binding domains will have a similar HTH fold with a DNA docking mode analogous to the Tc3/paired/Hin recombinase/λ repressor family. Detailed predictions on the mode of interaction with DNA are, however, not possible. In fact, analysis of the C.elegans Tc1 transposon DNA sequence would lead to a prediction that any potential complementarity of protein and DNA in the case of Tc1 should give a very different complex because the DNA sequence is very different from that in Tc3. However, it is likely that again a combination of sequence‐dependent conformability with base‐specific recognition plays an important role in DNA recognition by other members of the Tc1/mariner family.
Thus, the structure of the Tc3 transposon–transposase complex gives a good explanation of the specificity of recognition of Tc3, but does not help to predict other transposon–transposase complexes in detail.
The synaptic complex. The complete Tc3 transposase molecule, as outlined in Figure 1, consists of a bipartite DNA‐binding domain and a catalytic domain. Using the similarity to the paired domain structure, we can make some predictions about the second half of the bipartite DNA‐binding domain of Tc3A. It is likely that the second half of the domain also contains an HTH motif, similar to that in paired, according to the weak sequence homology between the Tc1/mariner family and the paired domain. This prediction fits well with secondary structure prediction with PHDsec. In agreement with footprinting data, this second domain will probably bind to the DNA closer to the cleavage site, roughly in the same location as in the paired domain, since the C‐terminal flexible region of Tc3A‐N (53–65) runs in an equivalent direction to the flexible connecting loop in the paired domain. Nevertheless, the paired domain structure cannot make any prediction about the DNA‐binding mode of the second HTH domain in Tc3A, since this region does not interact with DNA in paired form. Another family in which bipartite HTH domains occur is the family of eukaryotic homeodomain proteins. There the two HTH domains can interact with DNA in a variety of ways (for a review, see Tullius, 1995). It is conceivable that a similar variety will be seen in the Tc3/paired family.
To carry out the transposition reaction, the transposase must first recognize and synapse the two transposon termini before cleaving both the ends. The cleavage site in the transposon DNA has to be positioned properly within the active site of the catalytic domain of the transposase. Maybe the non‐linear conformation of the transposon DNA, as seen in the crystal structure, plays a role in bringing the cleavage site to the active site on the catalytic domain of the protein. However, given the importance of the DNA deformation for specific recognition between Tc3A and its transposon, the DNA conformation could be quite different in other Tc1/mariner elements. Of the two DNA bends, the bend at the TATA sequence, which is both closer to the cleavage site and somewhat conserved in C.elegans Tc1 DNA, is more likely to play a role than the bending in the G:C stretch which is not conserved in Tc1 at all.
For actual synapse formation, multimerization of the protein is likely to be important. On the basis of sequence comparisons of several fish Tc1‐like transposable elements, a conserved, so‐called leucine zipper motif has been indicated (Ivics et al., 1996), with a suggested importance for dimerization/oligomerization of the proteins. However, the sequences referred to do not fold into a leucine zipper in the three‐dimensional structure of Tc3A, but form part of the hydrophobic core of the protein, spread over the three helices.
In the crystal structure, we observe a crystallographic dimer of the Tc3A DNA‐binding domains, in which each monomer binds one DNA molecule. It is plausible that this dimer provides the first glimpse of the way in which the Tc1 family brings the two transposon DNA ends together. Once the two ends are synapsed with a multimer of the protein, cleavage of the transposon can take place. In bacteriophage Mu, this DNA cleavage occurs in trans, i.e. the monomer that is bound to a DNA strand cleaves the opposite strand (Aldaz et al., 1996; Savilahti and Muzuuchi, 1996; Yang et al., 1996). In the crystallographic Tc3 dimer, the two DNA monomers are found in a parallel orientation, with the C‐terminal regions of the Tc3A‐N pointing in the same direction, indicating roughly where the second DNA recognition and catalytic domains will be located. It is possible that the catalytic domains of Tc3A also act in trans, with each monomer cutting the opposite strand of the DNA, thus ensuring that simultaneous cleavage of the two transposon ends takes place.
Materials and methods
Expression and purification
The N‐terminal 65 amino acids of Tc3A with a C‐terminal His tag were expressed in Escherichia coli from the pET3c‐derived vector pRP1200. This vector was constructed from pSDC328 (Colloms et al., 1994), by cloning the annealed oligonucleotides RK1 (5′ CCATCACCATCACCATCACTAGA 3′) and RK2 (5′ AGCTTCTAGTGATGGTGATGGTGATGG 3′) into the SpeI–HindIII site to add the C‐terminal His tag. Cultures of E.coli strain BL21(DE3)pLysS transformed with pRP1200 were grown at 37°C and protein expression was induced (at OD600 ≈ 0.6) by the addition of 0.4 mM isopropyl‐β‐d‐thiogalactopyranoside (IPTG) 4 h prior to harvesting. The cells were resuspended in lysis buffer (6 M guanidinium chloride, 100 mM NaCl, 50 mM Tris pH 7.5, 1 mM β‐mercaptoethanol) and sonicated. Insoluble material was removed by centrifugation.
The supernatant of the lysate, to which 5 mM imidazole was added, was allowed to bind to Ni‐NTA resin (Qiagen) overnight. The resin was loaded into a column and washed with five column volumes of lysis buffer containing 20 mM imidazole. The protein was eluted with 250 mM imidazole in lysis buffer and 1% β‐mercaptoethanol was added to the pooled fractions. The protein was then reduced fully, denatured at 37°C for 2 h and subsequently purified further on a gel filtration column (Superdex 75, High load 26/60, Pharmacia), in lysis buffer with 10 mM β‐mercaptoethanol. The protein was renatured by dialyzing against 50 mM sodium acetate pH 4.8, 0.4 M NaCl, 10 mM β‐mercaptoethanol, 10 mM EDTA at 4°C. The renatured protein was loaded on a cation exchange column (S‐Sepharose Fast Flow, Pharmacia), washed with three column volumes of 0.6 M NaCl and eluted with 1.5 M NaCl. The pure protein (as judged from an SDS–PAGE analysis) was dialyzed against 20 mM sodium acetate pH 4.8, 0.1 M NaCl, 10 mM β‐mercaptoethanol and concentrated to 5–7 mg/ml. The renatured protein gives a single peak on an analytical gel filtration column (Superdex 75, Pharmacia) in 20 mM sodium acetate pH 4.8, 0.2 M NaCl, 10 mM β‐mercaptoethanol. This peak corresponds to a mol. wt of 25 kDa (the size of the monomer is 8.3 kDa). N‐terminal sequencing of the purified protein showed that the N‐terminal methionine was absent.
The 20‐ and 21‐DNA oligomers (Figure 1) were synthesized on a 1 μmol scale (Isogen) and purified on a 6 ml Resource Q column (Pharmacia) in 10 mM NaOH with an NaCl gradient. Initially, heavy precipitate formation upon mixing of protein and DNA occurred. Varying the salt concentration, pH or buffer did not affect precipitation, but variations in dilution and the ratio of DNA and protein prevented complete precipitation. The oligomers were annealed and mixed in a 1:1 molar ratio with the purified protein. Prior to mixing, the protein and DNA were diluted in 10 mM HEPES pH 7.5, 50 mM NaCl, 10 mM β‐mercaptoethanol. The complex was 50‐fold reconcentrated in Centricons to an OD260 nm, 1 cm of 70 (4.5 mg/ml complex).
Crystals were grown by the vapor diffusion method, from various precipitants and with different DNA oligomers, but most diffracted poorly. The best crystals (0.2×0.2×0.1 mm3 in size) used in this study grew within a week in hanging drops at 4°C. The reservoir contained 100 mM NaCl, 20 mM CaCl2, 13–15% MPD and 10 mM dithiothreitol (DDT) buffered with 50 mM sodium acetate at pH 5.5. The drop contained 1 μl of reservoir and 1 μl of 4.5 mg/ml protein–DNA complex stock solution prior to equilibrium.
Data collection and processing
A diffraction data set of a native crystal was collected at 4°C using synchrotron radiation in beamline X11 at the EMBL Outstation at DESY in Hamburg, using a MAResearch image plate. A second crystal was soaked in cryo‐protectant (mother liquor with the MPD concentration increased to 45%) for several minutes at 4°C and flash frozen in a nitrogen stream. A native data set was collected at −160°C in beamline BW7A. All data sets are anisotropic. While the resolution limit along the direction of the crystallographic b‐axis is 2.85 Å and 2.45 Å for the 4 and −160°C native data sets respectively, the diffraction limits along the a‐ and c‐direction are ∼0.5 Å lower. The intensities were integrated using the program DENZO (Otwinowski, 1993). Regions with absent reflections in the a‐ and c‐direction due to the anisotropy were excluded from integration in the cryo data set. Reflections were scaled and merged using the program SCALEPACK (Otwinowski, 1993). Intensities were converted to amplitudes using the TRUNCATE program (French and Wilson, 1978). The statistics are given in Table I.
Heavy atom derivatives and phasing
Heavy atom derivatives were prepared by soaking crystals in solutions containing 1 mM methylmercury chloride or 3 mM mercury thiocyanate in mother liquor. A DNA oligomer in which one thymine (T9) was substituted for 5‐iodouracil was also used to obtain crystals. Data sets for the first two derivatives were collected on the home‐source (a Nonius FR591 rotating anode generator) with a Dip2000 image plate and for iodinated DNA in beamline X31 at the EMBL Outstation in Hamburg. Heavy atom‐binding sites were determined from the difference Patterson map and confirmed in the anomalous difference Patterson map. The 4°C native data set was used as the reference set throughout phasing. All derivatives contained only one heavy atom per asymmetric unit. The multiple isomorphous replacement method with anomalous data (MIRAS) was used. MIRAS phases were calculated using the program PHASES (Furey and Swaminathan, 1990). Reflections with F<3σ(F) were rejected from phase refinement. The data were cut off at the resolution (see Table I) where the phasing power dropped below 1.0. Heavy atom refinement statistics are given in Table I. The solvent‐flattening technique as implemented in PHASES was applied to improve the initial MIRAS map.
Model building and refinement
Most of the DNA and two α‐helices of the protein were recognized in the resulting electron density map and built into the electron density using the program O (Jones et al., 1991; Kleywegt and Jones, 1996). This model was translated to the slightly different unit cell of the frozen crystal. Five percent of the data were set aside as a test set (Brünger, 1992b). Rigid body and simulated annealing refinement against the cryo native data was performed with the XPLOR program (Brünger, 1992a). The Rfree dropped from 55 to 44% with an R‐factor of 33%. Still, the third, C‐terminal, α‐helix and most of the side chains were not recognizable in a 2Fo−Fc or SIGMAA‐weighted (Read, 1986) electron density map. The partial model was refined with the program ARP (Lamzin and Wilson, 1993, 1997) using loose stereochemical restraints for the DNA and the existing protein region. In that procedure, reciprocal maximum‐likelihood refinement using REFMAC (Murshudov et al., 1997) was followed by automatic model update by ARP, for several cycles. ARP was instructed to introduce atoms as close as 1.2 Å from the existing protein or DNA atoms, to build missing model parts. After 50 cycles, the Rfree dropped to 34%. The 2mFo−DFc SIGMAA‐weighted electron density map showed clear density for most missing parts. This map was used to add the side chains and C‐terminal residues to the XPLOR model. Amino acid residues 2–52 and the complete DNA were built and refined using simulated annealing (Rfree = 32.2%). Final refinement steps were performed with the refinement program TNT (Tronrud, 1996), which resulted in a final model with an R‐factor of 23.4% and an Rfree of 31.8% at 2.45 Å (refinement statistics are given in Table I).
The R‐factor and Rfree (Brünger, 1992b) are relatively high. This could be caused by the presence of the flexible C‐terminus and/or the incompleteness of the data due to anisotropy. The model has good stereochemistry (Table II). The PROCHECK program (Morris et al., 1992) showed that the main‐chain dihedral angles for the majority of the residues lie in the most favorable regions of the Ramachandran plot and five in the additional allowed regions. The mean B‐factor for all atoms has a high value of 48 Å2 (protein atoms 41 Å2, DNA atoms 51 Å2, solvent atoms 53 Å2). The predicted B‐factor from a Wilson plot (1949) is even higher, 58 Å2. A stereo view of representative electron density is shown in Figure 7. The electron density of a 2Fo–Fc map contoured at 1σ covers the model almost completely. For a few side chains (with high B‐factors) on the surface of the protein, no electron density is present, due to flexibility. The electron density at the end of the side chain of Arg3 is not very clear. This puts some doubt on the relative positions of Pro2 and Arg3, but refinement of the two alternative positions gave the best R‐factors and stereochemistry with the current positions. The conformation of the last visible amino acid at the C‐terminus, Ser52, is not completely clear in the electron density map. Two bases at one end of the double‐stranded DNA oligomer are flipped outward. These do not fit in the electron density completely. The rest of the model fits very well in the electron density. The coordinates are submitted to the Brookhaven Protein Data Bank (1tc3).
We thank Dr Alexei Teplyakov for assisting with data collection, the members of the protein structure group at the NKI and Luca Jovine (MRC Laboratory of Molecular Biology, Cambridge) for helpful discussions. R.F.K. is supported by grant 700‐35‐210 from NWO/SON. A.P. is an EMBO fellow, ALTF‐215/1995. We thank the European Union for support of the work at EMBL Hamburg through the HCMP Access to Large Scale Facilities grant, Contract Number CHGE‐CT93‐0040.
- Copyright © 1997 European Molecular Biology Organization