Chemical trapping of ternary complexes of human immunodeficiency virus type 1 integrase, divalent metal, and DNA substrates containing an abasic site. Implications for the role of lysine 136 in DNA binding.

We report a novel assay for monitoring the DNA binding of human immunodeficiency virus type 1 (HIV-1) integrase and the effect of cofactors and inhibitors. The assay uses depurinated oligonucleotides that can form a Schiff base between the aldehydic abasic site and a nearby enzyme lysine ε-amino group which can subsequently be trapped by reduction with sodium borohydride. Chemically depurinated duplex substrates representing the U5 end of the HIV-1 DNA were initially used. We next substituted an enzymatically generated abasic site for each of 10 nucleotides normally present in a 21-mer duplex oligonucleotide representing the U5 end of the HIV-1 DNA. Using HIV-1, HIV-2, or simian immunodeficiency virus integrases, the amount of covalent enzyme-DNA complex trapped decreased as the abasic site was moved away from the conserved CA dinucleotide. The enzyme-DNA complexes formed in the presence of manganese were not reversed by subsequent addition of EDTA, indicating that the divalent metal required for integrase catalysis is tightly bound in a ternary enzyme-metal-DNA complex. Both the N- and C-terminal domains of integrase contributed to efficient DNA binding, and mutation of Lys-136 significantly reduced Schiff base formation, implicating this residue in viral DNA binding.

To analyze the extent of strand transfer using the "precleaved" substrate, AE118S and deoxyuridine analogs (see Fig. 3A) were 5Ј-endlabeled, annealed to AE117, and column-purified as above.
To analyze the choice of nucleophile for the 3Ј-processing reaction, AE118 and deoxyuridine analogs (see Fig. 3A) were 3Ј-end-labeled using ␣-[ 32 P]cordycepin triphosphate (DuPont NEN) and terminal transferase (Boehringer Mannheim). The transferase was heat-inactivated and RM22M was added to the same final concentration. The mixture was heated at 95°C, allowed to cool slowly to room temperature, and run on a G-25 spin column as before.
Chemical and Enzymatic Depurination-Duplex oligonucleotides were depurinated by incubation of 35 l of end-labeled duplex (500 nM stock concentration) with an equal volume of formic acid (50% final concentration) for 20 min at 40°C. The reaction was then dried under vacuum, redissolved in 150 l of water, dried under vacuum again, and redissolved in 35 l of water.
Duplex oligonucleotide substrates containing a single enzymatically generated abasic site were created as follows. Analogs of AE118 (see sequence above and in Fig. 3A) were synthesized so that one deoxyuridine replaced each of the wild-type nucleotides in this strand. For example, substrates Ϫ1 and Ϫ11 (see Fig. 3A) had the sequences 5Ј-GTGTGGAAAATCTCTAGCUGT-3Ј and 5Ј-GTGTGGAAUATCTCTAG-CAGT-3Ј, respectively. Each of these single strands was then radiolabeled and annealed to the complementary strand AE117 as described above. The uracil was removed from duplex oligonucleotides containing deoxyuridine by incubation of 40 l of end-labeled duplex (500 nM stock concentration) with 1 unit of uracil DNA glycosylase (Life Technologies, Inc.) for 2 h at 30°C. The reaction was then loaded on a G-25 Sephadex quick spin column to remove the unincorporated label and the uracil.
The extent of AP site formation was determined by incubation of 0.5 M of the radiolabeled AP site-containing DNA with 166 mM sodium hydroxide for 30 min at 30°C. The extent of cleavage of the oligonucleotide by ␤and ␦-elimination reactions was quantitated and determined to be 100%, implying that all of the uracil from the substrate had been excised.
Integrase Proteins-Recombinant HIV-1 integrase was purified as described elsewhere (13). Purified recombinant wild-type SIV integrase was a generous gift of Drs. R. Craigie and A. Hickman, Laboratory of Molecular Biology, NIDDK, NIH, Bethesda, MD. A plasmid encoding the HIV-2 integrase was generously provided by Dr. R. H. A. Plasterk (Netherlands Cancer Institute) and purified as described previously (14).
Schiff Base Formation and Chemical Trapping-A solution of sodium borohydride (1 M stock concentration, unless otherwise specified) was prepared fresh prior to the start of the experiment. Integrase was incubated with the oligonucleotide containing the chemically or enzymatically generated abasic site in reaction buffer (as described in "3Ј-Processing and Strand Transfer Assays") for 2 min at room temperature (unless otherwise specified). Sodium borohydride was then added (0.1 M final concentration, unless otherwise specified), and the reaction was continued for an additional 5 min. An equal volume (16 l) of 2 ϫ SDS-PAGE buffer (100 mM Tris, pH 6.8, 4% 2-mercaptoethanol, 4% SDS, 0.2% bromphenol blue, 20% glycerol) was added to each reaction, and the reaction was heated at 95°C for 3 min prior to loading a 20-l aliquot onto a 12% SDS-polyacrylamide gel. The gel was run at 120 V for 1.5 h, dried, and exposed in a PhosphorImager cassette. For inhibition of DNA binding experiments, integrase (200 nM) was preincubated with the inhibitor (at the indicated concentration) for 30 min at 30°C prior to the subsequent addition of the radiolabeled viral DNA substrate (20 nM) and borohydride. Gels were analyzed using a Molecular Dynamics PhosphorImager.
UV Cross-linking Experiments-The method used has been described by Yoshinaga et al. (15). Briefly, integrase was incubated with substrate in reaction buffer as above for 5 min at 30°C. Reactions were then irradiated with a UV transilluminator (254-nm wavelength) from 3 cm above (2.4 milliwatts/cm 2 ) at room temperature for 10 min. An equal volume (16 l) of 2 ϫ SDS-PAGE buffer was added, and an aliquot was electrophoresed as described above.

Trapping a Schiff Base Formed between HIV-1 Integrase and a Chemically Depurinated Duplex U5 LTR Substrate-
The acid-catalyzed depurination of DNA results in the formation of abasic sites (16) which are predominantly a mixture of the ␣and ␤-anomers of the hemiacetal (17) (Fig. 1). These anomers exist in equilibrium with the acyclic aldehydic form, which accounts for approximately 1% of abasic sites (18). Because abasic sites do not significantly perturb the structure of B-DNA (19,20), we reasoned that introduction of one such lesion into a duplex U5 LTR oligonucleotide may not alter its ability to act as a substrate for HIV-1 integrase. Furthermore, HIV-1 integrase has previously been shown to form a Schiff base (an imine linkage) between oxidized ATP and, presumably, a lysine ⑀-amino group in the proximity of this nucleotide binding site (21). We therefore tested the ability of several U5 LTR duplexes of increasing length to form a Schiff base upon integrase binding (Fig. 1). The Schiff base could be detected after reduction of the DNA-enzyme complex with sodium borohydride to stabilize the otherwise labile imine (Fig. 1). The resulting covalent enzyme-DNA complex was detected by SDS-PAGE analysis and PhosphorImager visualization.
The sequences of the duplexes used are shown in Fig. 2A. As seen in Fig. 2B, duplex substrates which were not depurinated (both panels, lanes 2) or which were depurinated but not subjected to borohydride reduction (both panels, lanes 5 and 8) could not form a detectable DNA-enzyme complex. Depurinated substrates generated a DNA-enzyme complex which was detected after borohydride reduction (both panels, lanes 6 and 9). The ability to form a Schiff base with the abasic site was observed whether the top or bottom strand was depurinated (both panels, compare lanes 6 and 9). If the aldehydic abasic site was reduced to the primary alcohol (with borohydride) prior to the addition of integrase, a DNA-enzyme complex could not be detected (both panels, lanes 10 and 11), as would be expected from the loss of an electrophilic carbonyl. With any of the three substrates, the DNA-enzyme complex demonstrated the expected mobility of a complex of the DNA and an enzyme monomer (39 and 43 kDa for the left and right panels, respectively) and had a similar mobility as a DNA-enzyme complex generated by UV cross-linking (11, 15) (both panels, lanes 12). A DNA-enzyme complex was also obtained when a 50-mer depurinated duplex oligonucleotide substrate was used (data not shown). After treatment with borohydride, a very low level of DNA-enzyme complex was detectable with control substrates which were not depurinated (both panels, lanes 3), presumably due to a low level of spontaneous depurination in DNA (16).
Recent photocross-linking studies have shown that binding of full-length wild-type HIV-1 integrase does not require a metal ion (11,15). Furthermore, both 3Ј-processing and strand transfer can be efficiently performed by HIV-1 integrase using either manganese or magnesium (22,23). We therefore determined the extent of Schiff base formation using substrates of increasing length in the absence or presence of divalent metal ion. As seen in  7). Similarly, a DNA-enzyme complex was also observed with either the 35-or 50-mer duplex oligonucleotide substrates (middle and right panels). Interestingly, the extent of DNA-enzyme complex formation was less dependent on the presence of divalent metal ion when the longer 35-or 50-mer duplexes were used (all panels, compare lanes 3 and 5). We conclude that, with the 21-mer substrate, the enhanced level of DNA binding by integrase in the presence of manganese versus magnesium parallels the enhanced catalytic activity observed in the presence of manganese. However, as the length of the substrate is increased, the ability of integrase to efficiently perform its 3Ј-processing activity in magnesium is also increased (24), consistent with results from our DNA binding assay (Fig. 2C).
Enzymatically Generated Abasic Sites Allow Fine Mapping of Optimal Sites for Schiff Base Formation-In order to determine whether the extent of Schiff base formation depended on the specific location of the abasic site in the DNA, we created a set of 10 oligonucleotides, each having one enzymatically generated abasic site at a known location. The sequence of the wild-type 21-mer U5 LTR duplex, the scissile phosphodiester bond, and the position where an abasic site was substituted for each of the nucleotides are shown in Fig. 3A for both the blunt-ended and 3Ј-processed substrates. We have previously described and used these oligonucleotides to determine the effect of uracil misincorporation and a missing base on the catalytic activity of HIV-1 integrase (25). In addition, we created three duplex substrates, which were "precleaved" to represent the product of the 3Ј-processing reaction, each of which contained an abasic site at a known location.
As seen in Fig. 3B, when either HIV-1, HIV-2, or SIV integrase was incubated with each of these substrates, a DNAenzyme complex was detected after reduction with borohydride (odd numbered lanes). Uracil-DNA glycosylase cannot excise uracil from either the ultimate or penultimate bases near the 3Ј-end of a duplex (26), as in the case of position 1 (where a uracil was substituted for the guanine). However, the very low level of DNA-enzyme complex detected when this substrate was used reflects the low abundance of abasic sites present in this oligonucleotide (see Fig. 2, lane 3). The level of Schiff base formation was quantitated (in terms of pixel units), and the correlation between the extent of DNA binding and location of the abasic site was plotted in Fig. 3C. Interestingly, the amount of complex decreased as the abasic site was moved further away from the conserved CA dinucleotide (from position Ϫ1 to around Ϫ5 to Ϫ7) until a point where it no longer decreased or actually increased in all cases (at Ϫ5 or Ϫ7). This decrease was observed whether the blunt-ended or the "precleaved" duplex substrates were used (Fig. 3B, compare lanes 1 through 19 and  20 through 25).
A kinetic study of Schiff base formation showed that a timedependent increase in the extent of Schiff base formation was evident using each of the oligonucleotide substrates. Although substrates Ϫ5 through Ϫ11 displayed only a small increase in complex formation over time, substrates Ϫ2 and Ϫ3 P exhibited significant increases in the amount of complex formation (data not shown).
In our efforts to optimize the binding conditions, we also investigated the effect of varying the substrate DNA concentration. We found that the trend observed in Fig. 3 was reproducible over a DNA oligonucleotide concentration range of 0.2 to 75 nM (data not shown).
Detection of Metal Ion-Integrase Complexes-The metal ion specificity for Schiff base formation with each of the abasic site-containing oligonucleotides was also tested in an attempt to probe enzyme interactions both close to and further away from the conserved CA dinucleotide. Recently, the crystal structure of ASV integrase complexed with a single manganese ion has been solved (27). This ion is coordinated to the carboxylate groups of the active site aspartates of integrase and uses water to complete its octahedral coordination. Possibilities still exist for the DNA substrate to complete the coordination sphere of the metal or for the DNA substrate to bring a second metal into the active site.
As seen in Fig. 4A, a low level of DNA-enzyme complex was detected with all the substrates tested in the presence of EDTA, but the amount of complex formed was strongly enhanced by the presence of divalent cation, consistent with results obtained from the chemically depurinated oligonucleotide substrates (see Fig. 2C). A difference was observed, however, in that the metal ion specificity for Schiff base formation did not differ significantly depending on the location of the abasic site, with manganese and magnesium generating comparable levels of Schiff base with all substrates tested (Fig. 4, A and B, compare conditions C and D and Fig. 2C, left panel,  compare lanes 5 and 7).
The role of the metal ion in integrase structure and function has recently been investigated. Besides its role in catalysis, the FIG. 3. Schiff base formation with U5 oligonucleotides containing a single enzymatically generated abasic site (see Fig. 1). A, sequence of the 21mer duplex oligonucleotides (blunt-ended or 3Ј-processed) corresponding to the U5 end of the HIV-1 LTR. The scissile bond in the 3Ј-processing reaction is shown by the arrow, and nucleotides upstream and downstream of this bond are denoted by negative and positive numbers, respectively. Each base which is annotated with a number was replaced by an abasic site at that position. Thus, a set of 11 oligonucleotides were generated, each containing one abasic site at a known location. B, PhosphorImager picture of a typical experiment showing the levels of Schiff base formation using either HIV-1, HIV-2, or SIV integrase and the substrates shown in A. Molecular mass markers (in kDa) are shown to the right of each gel. The absence or presence of sodium borohydride (NaBH 4 ) to each reaction is denoted by "Ϫ" or "ϩ," respectively. The substrate used in each set of reactions (Ϫ and ϩ NaBH 4 ) is shown at the top of the panel. metal ion (either manganese or magnesium) is required to promote stable complex formation (28 -30), possibly by promoting specific interactions between integrase protomers, leading to protein multimerization (30,31). We tested the stability of the metal ion-promoted DNA binding activity of HIV-1 integrase by adding EDTA prior to (Fig. 4, C and D) or after (Fig.  4, E and F) the addition of DNA. As seen in Fig. 4C, an integrase-DNA complex was detected in the presence of either EDTA (lane 2) or manganese (lane 8), although preincubation of the enzyme in manganese greatly enhanced the level of DNA binding, consistent with data from Fig. 4, A and B. Addition of increasing concentrations of manganese after preincubation of integrase in 2 mM EDTA resulted in only a slight increase in DNA binding (Fig. 4, C, lanes 3-7, and D). Increasing concentrations of EDTA after preincubation of integrase in 2 mM MnCl 2 resulted in an EDTA concentration-dependent reduction in the level of DNA binding (Fig. 4, C, lanes 9 -13, and D).
However, when each of these additions was performed after the addition of DNA (i.e. after a ternary integrase-metal-DNA complex had formed), different results were obtained (Fig. 4, E and F). For example, addition of increasing concentrations of manganese after preincubation of integrase in 2 mM EDTA resulted in approximately 10-fold increases in the DNA binding at concentrations equimolar to or in excess of the EDTA concentration (Fig. 4, E, lanes 5 and 6, and F). Addition of EDTA after preincubation of integrase in 2 mM MnCl 2 did not result in a significant reduction in the level of DNA binding (Fig. 4, E,  lanes 9 -13, and F). We conclude that preformed binary integrase-manganese complexes can be dissociated by EDTA, resulting in a reduction in the level of DNA binding by the enzyme, but that preformed ternary integrase-manganese-DNA complexes are stable in the presence of EDTA.
Binding of HIV-1 Integrase to Different DNA Substrates-HIV-1 integrase can bind both HIV-1 LTR and non-LTR DNA (3)(4)(5)(6), consistent with its reaction mechanism, in which a viral DNA strand is inserted into a target DNA strand and with genetic data demonstrating lack of strict target site selectivity in the integration reaction (32,33).
The level of Schiff base formation with substrates Ϫ2 and Ϫ11 (see Fig. 3A) was compared to that obtained with either a mutated form of Ϫ11 (Ϫ11 GCA3 CGT ) or an heterologous oligonucleotide in which an abasic site was generated at an analogous location as in the Ϫ2 oligonucleotide (Ϫ2 HET ). It has previously been demonstrated that mutation of the GCA trinucleotide to a CGT (as in Ϫ11 GCA3 CGT ) reduces 3Ј-processing activity to about 7% (34). The same study showed that the heterologous oligonucleotide substrate (denoted Ϫ2 HET ) was not processed by integrase. Integrase was able to form a Schiff base with all of these substrates (data not shown). We also found that HIV-1 integrase was able to form a Schiff base with single strands (where only the processed strand was present) of the substrate shown in Fig. 3A, consistent with previous results from UV cross-linking (11). We conclude that, in a 21-mer duplex substrate (e.g. those in Fig. 3A), integrase does not demonstrate highly selective binding to one sequence over others, consistent with binding of integrase to a variety of chromosomal sequences (32).
We also tested the ability of HIV-1 integrase to form a Schiff base with a shorter (16-mer) version of the 21-mer Ϫ2 duplex (Fig. 5A). This substrate has previously been shown to support both the 3Ј-processing and strand transfer activities of HIV-1 integrase (34) (see Fig. 5D). The level of DNA binding with this substrate was compared to that obtained with hairpin versions of this duplex where a two base loop (TT) was present either proximal or distal to the processing site (Fig. 5A). The results are shown in Fig. 5, B and C. There was a lower level of binding to (Fig. 5, B and C) and processing of (Fig. 5D) the PL substrate than the LIN or DL substrates. These data suggest a greater affinity for the processed end of the LTR, consistent with data in Fig. 3.
Schiff Base Formation with Site-directed and Deletion Mutants of HIV-1 Integrase-UV cross-linking studies have shown that residues 213-266 of the C terminus of HIV-1 integrase are required for efficient cross-linking to the linear duplex U5 oligonucleotide in the absence of metal ion (11). In fact, a deletion mutant containing only residues 215-270 can bind to DNA in the presence of 2 mM EDTA (9). However, IN 1-212 can bind to the linear duplex in the presence of MnCl 2 at 2.56 M protein concentration (11). We analyzed the binding of sitedirected and deletion mutants of HIV-1 integrase to three linear duplex oligonucleotides containing an abasic site (see Fig. 3A). As seen in Fig. 6A, HIV-1 integrase containing the F185K/C280S mutations, which make the protein more soluble without compromising catalytic activity (13), was able to bind to the DNA substrates (lanes 3, 10, and 17) as well as wild-type integrase (lanes 2, 9, and 16). Elimination of the zinc finger, either by site-directed mutagenesis of the two histidines in this domain (lanes 4, 11, and 18) or by deletion of the N terminus (lanes 6, 13, and 20) did not suppress the ability to form a DNA-enzyme complex in the presence of MnCl 2 . As expected, the IN 50 -288 protein ( lanes 5, 12, and 19) was able to bind to the oligonucleotide substrate probably due to its DNA-binding domain. IN 1-55 , containing only the N terminus of the protein, displayed a low level of DNA-protein complex of the expected molecular weight (lanes 7, 14, and 21) but also generated a DNA-protein complex having a large molecular weight of unknown identity.
Consistent with earlier results from UV cross-linking (11), an IN 50 -212 deletion mutant containing only the central catalytic domain was not able to form a Schiff base with the linear duplex oligonucleotide substrates containing an abasic site (Fig. 6B, lanes 4 and 8). This lack of binding, however, was not due to the lack of an appropriately positioned amino group in the central domain because an IN 50 -212 mutant containing a histidine tag (which contains no lysine residues but allows the deletion mutant to bind to linear DNA) was able to bind to  2-7, 9 -14, and 16 -21, with integrase at a final concentration of 5, 10, 20, 40, 80, and 160 nM, respectively. C, graph of the extent of Schiff base formation (in pixel units) versus concentration of integrase for each oligonucleotide. Data were derived from the experiment shown in B. D, extents of 3Ј-processing and strand transfer using the oligonucleotides in A except that the abasic site was replaced by the wild-type nucleotide C usually present at that position. Lanes 1-3, substrate was a linear 16-mer duplex oligonucleotide representing the U5 end of the LTR; lanes 4 -6, substrate was a hairpin 16-mer duplex oligonucleotide representing the U5 end of the LTR with a two base (TT) loop distal to the processing site; lanes 7-9, substrate was a hairpin 16-mer duplex oligonucleotide representing the U5 end of the LTR with a two base (TT) loop proximal to the processing site. Lanes 1, 4, and 7, DNA alone; lanes 2, 3, 5, and 6, and 8 and 9, with integrase at a final concentration of 80 and 27 nM, respectively. these same oligonucleotide substrates (Fig. 6B, lanes 3 and 7).
Site-directed mutagenesis and sequence alignment have identified three amino acid residues in the catalytic core, which are conserved among all retroviral integrases (35) and are critical for activity (36). These are Asp-64, Asp-116, and Glu-152. IN D116N has previously been shown to bind to the linear U5 duplex oligonucleotide by UV cross-linking (11). Consistent with those studies, we found that this mutant was able to bind to our abasic site-containing oligonucleotides and form a Schiff base (data not shown).
Some Possible Applications of the Assay-Structural insights into the interactions of integrase with its viral DNA substrate could be obtained using this assay. For example, a putative active site of HIV-1 integrase can be defined by the locations of the two catalytically essential aspartate residues at positions 64 and 116. The viral DNA binding site presumably overlaps with this active site. In the immediate vicinity of the active site are lysines at positions 103, 111, 127, 136, and 156. As an initial test to determine which of these lysines could be involved in DNA binding, we tested the ability of a site-directed mutant in which the lysine at position 136 was mutated to a glutamate to form a Schiff base with our library of oligonucleotides containing a single abasic site (see Fig. 3A). Four-to 10-fold higher levels of DNA binding were exhibited by wildtype HIV-1 integrase compared to the K136E mutant at protein concentrations of 80 nM and DNA concentrations of 25 nM (Fig.  7, A and B). Moreover, the pattern of Schiff base formation (with DNA substrate at 75 nM) was altered using this mutant compared to wild-type integrase (Fig. 7C). In Fig. 7C, the results using a 75 nM DNA substrate concentration were shown because of the higher level of DNA binding obtained with the K136E mutant at this concentration, but the general pattern was the same at 25 nM DNA substrate concentration. For example, the high levels of DNA binding observed with wildtype integrase when Schiff base formation occurred near the conserved CA dinucleotide were not observed with the K136E mutant. In contrast, the highest level of DNA binding with this mutant occurred with when a Schiff base was formed three bases 5Ј to the CA. These results suggest that lysine 136 plays a key role in Schiff base formation with substrates containing an abasic site within or 5Ј to the conserved CA dinucleotide. These results also suggest that multiple lysines are likely to be involved in the DNA binding (for example, interactions with substrate Ϫ5 ; Fig. 7C). Studies are in progress to define the lysines involved and, through this technique, elucidate the viral DNA binding site and residues involved in interactions with both the conserved CA dinucleotide and subterminal sequences.
We determined whether integrase which had formed a Schiff base with the abasic site-containing oligonucleotide substrate was still competent for catalysis, since the chemical crosslinking may be a fairly gentle perturbation of the integrase-DNA interaction. Wild-type integrase was incubated with the oligonucleotide substrate Ϫ11 (see Fig. 3A) prior to or concurrent with the addition of borohydride for increasing times. A DNA-enzyme complex was detected using both conditions (data not shown). Furthermore, both 3Ј-processing and strand transfer products were detected whether borohydride was added after or at the start of incubation (data not shown). No change was observed in the relative extents of glycerolysis, hydrolysis, and circular nucleotide formation (37,38) when HIV-1 integrase was chemically cross-linked to 3Ј-end labeled substrates at the start of the reaction versus after the reaction had proceeded (data not shown). We conclude that chemically cross- linked integrase is competent for catalysis.
Finally, we also tested whether our assay could be used to determine the inhibitory mechanisms of drugs. We preincubated HIV-1 integrase with either of two recently described inhibitors, the guanosine quartet structure formed by the oligonucleotide T30177 (39) 2 or a nonnucleoside inhibitor, the 4-hydroxycoumarin "butterfly" structure (41), prior to incubation with abasic site-containing oligonucleotide substrates (see Fig. 3A). We found that both inhibitors were able to inhibit the binding of integrase to the DNA substrates (data not shown), consistent with results from UV cross-linking (40,41). DISCUSSION The interactions of various proteins with nucleotide or polynucleotide substrates can be characterized by chemically generating an aldehyde which would be able to form an imine upon binding to the enzyme. This strategy was recently exploited using ATP which was first oxidized to the 2Ј,3Ј-dialdehyde. This modified nucleotide was then incubated with HIV-1 integrase and a Schiff base was detected upon reduction with NaCNBH 3 (21). The nucleotide binding site was further explored using pyridoxal phosphate, an inhibitor of HIV-1 integrase (42), which was also found to bind to integrase via a Schiff base (21). A similar strategy was used in studying the interactions of GTP (which was first oxidized to the 2Ј,3Јdialdehyde) with the ␣-subunit of the stimulatory G protein rG s␣-s (43). Abasic sites have also been created in polynucleotide substrates via chemical reagents and imine formation could then proceed by reaction of an appropriately positioned lysine with the aldehydic abasic site. Such a strategy has been used in defining the interactions between histones and DNA (44). This is the first report using this technique to probe a substrate DNA-enzyme interaction. This report describes the optimization of conditions and demonstrates the feasibility of this technique to gain further insight into the HIV-1 integrase DNA binding mechanism. Using oligonucleotide substrates of increasing length, we randomly introduced approximately one abasic site per duplex by chemical methods. This random "lesion mutagenesis" allowed us to rapidly screen for the formation of a Schiff base as well as for conditions where substrate length was important. We found that DNA binding through Schiff base formation was higher with manganese than magnesium with a 21mer duplex substrate. However, this differential level of complex formation was not observed when 35-or 50-mer duplex substrates were used (Fig. 2). These results are consistent with earlier findings describing the effect of substrate length on catalytic activity (24).
The enzymatic introduction of one abasic site per duplex at a known location allowed us to attempt a fine mapping of optimal sites of interaction with the DNA substrate using both wildtype and deletion mutants of HIV-1 integrase and to provide a basis for understanding results from substrate mutagenesis studies (33,34,(45)(46)(47). We found that optimal Schiff base formation occurred with substrates near the conserved CA dinucleotide (Fig. 3), consistent with results obtained from mutagenesis experiments which showed that these two nucleotides and those proximal to them are important for catalysis. These results are also consistent with our data showing that a loop proximal to the processing site diminishes the ability of integrase to bind to the substrate (Fig. 5). An alternative explanation for the wavelike dependence of Schiff base formation on the location of the abasic site could be that one lysine residue is close enough to the processing site such that it acts as the only nucleophile in the Schiff base formation for substrates Ϫ1 through Ϫ7. However, as the abasic site is moved further away from the processing site (e.g. substrates Ϫ5 through Ϫ7), this lysine residue becomes too far removed to efficiently form a Schiff base, and a second lysine residue may act as the nucleophile in the Schiff base formation with substrate Ϫ11 (and presumably with other substrates in which an abasic site is even further away from the processing site).
The presence of either manganese or magnesium greatly enhanced the level of Schiff base formation, although a low level of an enzyme-DNA complex was generated in the absence of divalent metal ion (Fig. 4), consistent with previous results from UV cross-linking (11). This increased binding in the presence of metal suggests that the metal participates in efficient DNA binding. A novel aspect of the current study is that preformed binary complexes containing integrase and bound divalent metal can be dissociated by the addition of EDTA (Fig.  4, C and D), resulting in a reduction in the level of DNA binding by the enzyme. These data suggest that the metal is loosely bound to the HIV-1 integrase in the absence of DNA, which is consistent with the fact that crystals of ASV integrase required soaking in high concentrations of either MnCl 2 (10 mM) or MgCl 2 (500 mM) in order to obtain clear electron-density maps of crystals of either divalent metal bound to the integrase (27). This is also consistent with the requirement of a high concentration of MnCl 2 (25 mM) for assembly of integrase-donor DNA complexes prior to the strand transfer reaction (30). However, the observation that addition of EDTA after the formation of the enzyme-metal-DNA complex failed to reverse the integrase-DNA complexes (Fig. 4, E and F) suggests the existence of stable ternary integrase-metal-DNA complexes. The existence of such a ternary complex is consistent with structural data from the DNA polymerase model of Beese and Steitz (40).
Site-directed as well as N-and C-terminal deletion mutants were able to form a Schiff base with the oligonucleotides containing an abasic site (Fig. 6), demonstrating that the lysine residue(s) which is involved in imine formation is located in the central catalytic domain. However, the presence of either the N or C terminus was required for efficient DNA binding because the deletion mutant IN 50 -212 was not able to bind to the linear duplex substrate.
Which lysine residue(s) forms the Schiff base with the abasic site in the DNA? As an initial attempt to address this question we tested a lysine to glutamate mutation at position 136 of HIV-1 integrase. This mutant protein demonstrates catalytic activity approximately equal to that of wild-type HIV-1 integrase in the protein concentration range of 500 nM and only slightly reduced activity at lower concentrations. 3 This mutant demonstrated significantly reduced although still detectable binding by our assay. At a protein concentration of 80 nM, there was a 4 -10-fold reduction (depending on the substrate used) in the level of DNA binding compared to wild-type HIV-1 integrase (Fig. 7). Moreover, the pattern in the level of Schiff base formation as a function of base sequence was altered using this mutant compared to the wild-type. Future mutation studies using this assay are planned to determine which lysine(s) residue comes into contact with the viral DNA substrate and to define the viral DNA binding site(s). In the absence of a cocrystal structure of integrase bound to its viral DNA substrate, such studies should provide valuable insight into interactions between residues on the integrase and the conserved CA dinucleotide and subterminal sequences.
In summary, we have described a novel and simple assay for monitoring the binding of HIV-1 integrase to its viral DNA substrate via the formation of a Schiff base with an enzymatically introduced abasic site. Requirements for Schiff base formation paralleled those for UV cross-linking and catalysis and may provide a basis for understanding substrate mutagenesis and catalysis. We provide experimental evidence for the stability of ternary integrase-metal-DNA complexes but not binary integrase-metal complexes in the presence of EDTA, consistent with a loose binding of metal ion by integrase in the absence of DNA. This novel approach may also provide insights into structural details of integrase interactions with its DNA substrate.