The metal ion-induced cooperative binding of HIV-1 integrase to DNA exhibits a marked preference for Mn(II) rather than Mg(II).

In this investigation, we examine the interaction between the human immunodeficiency virus type I integrase and oligonucleotides that reflect the sequences of the extreme termini of the viral long terminal repeats (LTRs). The results of gel filtration and a detailed binding density analysis indicate that the integrase binds to the LTR as a high-order oligomer at a density equivalent to 10 ± 0.8 integrase monomers per 21-base pair LTR. The corresponding binding isotherm displays a Hill coefficient of 2, suggesting that the binding mechanism involves the cooperative interaction between two oligomers. This interaction is quite stable, exhibiting a prolonged half-life (t≈ 13 h) in the presence of Mn cations. Complexes were less stable when formed with Mg (t≈ 1 h). The role of Mn appears to be in the induction of the protein-protein interactions that stabilize the bound complexes. In terms of the 3′-end processing of the LTR, similar catalytic rates (k ≈ 0.06 min) were obtained for the stable complex in the presence of either cation. Hence, the apparent preference observed for Mn in standard in vitro integration assays can be attributed entirely to the augmentation in the DNA binding affinity of the integrase.

The integration of retroviral DNA involves two separable DNA strand cleaving events at positions that consequently form the sites of union between the viral and host DNA. The first occurs in the cytoplasm of the infected cell and results in the removal of the two outermost bases from each 3Ј-end of the pre-integrated linear DNA genome (1)(2)(3). This cleavage activity occurs invariantly adjacent to a conserved CA dinucleotide, embedded within a short inverted repeat sequence present at the termini of the viral LTRs. The second strand breakage involves the coordination of the recessed viral 3Ј-OH as a nucleophile in an enzyme-catalyzed attack on the phosphodiester backbone of the host target DNA. Concurrent with the exposure of a host 5Ј-phosphate, a single, direct transesterification with the attacking viral 3Ј-OH end results in the insertion of the viral DNA at this site (4).
Whereas the integration reaction is mediated in vivo by a large nucleoprotein complex known to contain several retroviral proteins (5), both the 3Ј-end processing and strand transfer activities can be carried out in vitro using short duplex oligonucleotides that mimic one of the viral LTR termini and the purified retroviral integrase protein alone (6 -9). Hydrolysis of the viral substrates by the integrase is site-specific in the sense that it consistently cleaves 3Ј of the conserved CA dinucleotide, even when the 3Ј-LTR is artificially extended as singlestranded DNA (9). However, although several studies have confirmed that a level of sequence specificity is required for efficient cleavage and integration (9 -12), this is not reflected in a corresponding DNA binding preference for the cognate LTR (6,11,(13)(14)(15)(16)(17).
Several pertinent observations have been made with regard to the dominant nonspecific DNA binding properties of retroviral integrases. Deletion mapping studies have located the major determinant of the nonspecific DNA binding domain to a region within the C terminus of the integrase (18,19). Although poorly conserved and of variable length in sequence comparisons between retroviral integrases, this region is rich in basic residues (Arg, Lys) and may interact nonspecifically with the DNA via general electrostatic interactions with the phosphate backbone. A central catalytic core domain of the integrase, which possesses a highly conserved D,D-35-E motif (15,20), proposed to form the polynucleotidyl transfer catalytic site (21)(22)(23), also binds DNA but with an apparent requirement for branched structures that mimic the Y-type intermediate resulting from a single LTR integration event into a target DNA (17). The DNA binding properties of the core domain thus reflect the ability of this region to recognize and resolve such DNA structures by catalyzing the removal of the branch arm (LTR) and resealing the broken (target) strand in a process termed disintegration (24). It has been proposed that a second highly conserved region, a HHCC motif located near the N terminus of the integrase, may form a Zn 2ϩ -binding finger with similarities to those of the transcription factor IIIA-like DNA binding proteins (25). However, whereas this region has been shown to bind Zn 2ϩ in vitro (22), no evidence has been presented to define a direct interaction with the substrate DNA in binding studies. Rather, this region has been implicated in forming protein-protein interactions required for the divalent cation-induced oligomerization of the integrase (26). Interference with this motif, either by mutagenesis (16,21,23,27,28) or chemical modification (28), is detrimental to the 3Ј-end processing and strand transfer activities of the integrase, but such modifications do not appear to block disintegration or the nonspecific interaction with DNA. The observation that the modification of this domain prevents the formation of the stable interaction between the integrase and the LTR suggests that protein-protein interactions may be essential in the assembly of the functional complexes (26,29). The involvement of subunit interactions in the assembly of an active oligomeric form of the HIV-1 1 integrase has also been demonstrated by genetic (30) and in vitro complementation (31)(32)(33) studies of integrase mutants.
The refractory nature of this retroviral protein to analysis by conventional biochemical strategies, such as mobility shift and DNase footprinting techniques, has previously precluded a detailed, quantitative account of the biophysical properties of integrase nucleoprotein complexes. Recently, highly constrained solution conditions allowing a specific mobility shift to be detected between the HIV-1 integrase and the LTR were defined (34). Notably, these required low concentrations of the DNA substrate and the omission of the divalent cation essential for the catalytic processing of the LTR. In the present investigation, we describe the application of laser-mediated single-pulse UV cross-linking and nuclease digestion techniques to obtain a quantitative description of the stable binding of the HIV-1 integrase to the LTR, under solution conditions more closely reflecting those optimized for 3Ј-end processing and DNA integration assays. The implications of this data on the quaternary structure of the nucleoprotein complex and the influence of divalent cations on the stability of this interaction are examined.

Materials
Oligonucleotides used in this study were synthesized and purified by Genset (Paris). Snake venom phosphodiesterase I was purchased from Pharmacia (France). Factor Xa was obtained from Boehringer Mannheim (France). All chemicals used were analytical grade. MnCl 2 and MgCl 2 were obtained from Prolabo (Paris).

Methods
Expression and Purification of HIV-1 Integrase-The HIV-1 integrase encoded by the LAV-BRU isolate (35) was cloned and produced in a T7 expression system essentially as reported previously (36) but with the following modifications. To enable purification by metal chelation chromatography, the N terminus of the integrase was modified to bear the additional sequence M(H) 6 GSIEGR. The sequence IEGR was included to allow removal of the extraneous sequence by the factor Xa protease. Expression of the 6 ϫ His-tagged HIV-1 integrase construct was mediated in the Escherichia coli strain BL21(DE3) by induction with 0.5 mM isopropyl-1-thio-␤-D-galactopyranoside. The integrase was solubilized by resuspending the insoluble fraction of sonicated lysates in buffer A (40 mM Tris⅐HCl (pH 7.6), 1 M NaCl, 10 mM imidazole, 10 mM CHAPS) and stirring for several hours at 4°C. The solubilized protein was loaded onto a Ni 2ϩ -charged resin of chelating Sepharose (Hitrap Chelating, Pharmacia). Following extensive washing with buffer A containing 50 mM imidazole, the bound protein was eluted with 25 mM EDTA in buffer A. The eluted protein was diluted to 100 mM NaCl with buffer B (40 mM Tris⅐HCl (pH 7.6), 10 mM CHAPS) and loaded onto heparin-Sepharose (Hitrap Heparin, Pharmacia). The integrase was washed with 200 mM NaCl in buffer B and then eluted with integrase storage buffer (1 M NaCl in buffer B adjusted to 20% (v/v) glycerol and 5 mM dithiothreitol). Samples of HIV-1 integrase were snap frozen in a dry ice-ethanol bath prior to storage in small volumes at Ϫ70°C and thawed once only before use. For control experiments, the His-tag was removed by incubation with factor Xa. A 1:100 (wt:wt) ratio of factor Xa:integrase in 2-fold diluted integrase storage buffer (adjusted to 1 mM CaCl 2 ) was incubated for 2 h at 30°C. The cleaved protein was repurified from the cleaved His-tag sequence by a single passage through the Ni 2ϩ -chelating column and further purified from the factor Xa enzyme by heparin chromatography as described above. Protein micro-sequencing was employed to confirm the authenticity of the N-terminal integrase sequence produced in this manner. In practice, we were unable to distinguish any difference in the 3Ј-end processing activity between the His-tagged and factor Xa-processed forms of the recombinant protein.
Whereas it has been shown previously that the presence of an Nterminal His tag does influence the activity of N-terminal truncated derivatives of both HIV-1 (37) and RSV (38) integrases, this effect has further been demonstrated to be independent of the sequence of the extraneous peptide positioned in the deleted region of the protein (38).
We suspect that in the case of the core domain of the integrase, the presence of a His tag (or other peptide sequence) may play a critical role in maintaining the native structure and activity of this domain rather than demonstrate a direct influence of the His tag on nonspecific DNA binding per se, as has been suggested elsewhere (37). The experiments detailed herein, therefore, were conducted directly with the His-tagged derivative, employing the processed integrase only as a control where appropriate.
Exonuclease Protection Assay-DNA binding reactions were conducted under conditions stated above for 3Ј-end processing. After a 5-min pre-binding step, nuclease digestions were performed by the addition of snake venom phosphodiesterase I to a final concentration of 100 g/ml. This is equivalent to 0.8 units of phosphodiesterase I, where 1 unit is defined as the hydrolytic activity yielding 1 mol of p-nitrophenyl thymidine 5Ј-phosphate min Ϫ1 at 25°C. Reactions were continued for specified time intervals and terminated by the addition of a 95% formamide:20 mM EDTA solution to a 1:4 dilution and heating to 80°C. Products were analyzed on 20% denaturing polyacrylamide gels.
Laser-mediated UV Cross-linking-Rapid pulse UV cross-linking was performed using a high energy light source emitted by an Nd YAG laser (DCR-3G, Spectra Physics) as described previously (39). DNA binding reactions were performed as for 3Ј-end processing except that incubations were carried out at 21°C for 10 min prior to exposure to a single, 5-ns pulse of high energy monochromatic (266 nm) light. Adduct formation was monitored by SDS-polyacrylamide gel electrophoresis on 10 -15% gradient gels (Excelgel, Pharmacia). Since we observed that the quantum yield of adduct formation displayed saturation kinetics with the incident photon flux density, a power setting was selected that routinely gave us optimum cross-linking (40 mJ, 10 17 photons per pulse) such that increasing the beam intensity, or providing additional pulses, neither increased nor decreased the amount of adduct obtained from a single pulse. Titrations of fixed concentrations of the oligonucleotide were performed by varying the concentration of the integrase. The progressive fractional saturation of potential DNA-interacting sites was signaled by an increase in adduct formation, measured in terms of the percentage of the total DNA cross-linked (S obs ). Maximal cross-linking (S max ) was obtained under saturating concentrations of protein where all the DNA binding sites are occupied. Adduct formation may thus be represented simply as follows: where denotes the fractional saturation of the DNA lattice and c represents the quantum yield of adduct obtained at saturation of the lattice with ligand, a constant factor dictated by the photochemical reactivity of the oligonucleotide sequence occluded and the number and nature of protein-DNA contacts formed. The implicit requirement that c is independent of the value of , and hence remains constant throughout the binding isotherm, is a point that will be validated under "Results." Theory and Data Fitting Procedures-We adopt here a simple Hill formulation adapted to cases where more than one ligand molecule binds reversibly to a polymer in a strictly cooperative manner (we neglect, therefore, any cause of anti-cooperative behavior such as sequence preferences or binding configurations that may lead to lattice exclusion (40), as will be justified under "Results" and "Discussion"). The binding mechanism is then characterized by a single equilibrium between free integrase oligomers (I) in solution and a set of binding targets on each DNA molecule (D) so that the overall binding process leading to the higher order assembly of oligomers on the DNA may be written as follows: where, in this simplified case, the Hill coefficient, n, corresponds to the number of integrase oligomers that interact upon binding to the DNA substrate. To account for these cooperative interactions, the titration data, in terms of fractional saturation (), were fitted directly by computer-assisted least squares procedures to the appropriate isotherm corresponding to the above mechanism: where [D] T refers to the total DNA concentration and [I] 50 represents the concentration of free protein required to titrate the DNA to half saturation (for convenience, titrations are expressed in terms of integrase monomers, and consequently the equilibrium binding parameters obtained from the corresponding isotherms are also reported in monomeric units). For a first-order (independent site) binding mechanism (n ϭ 1), the isotherm reduces to a hyperbola, and the dissociation constant (K d ) is given by [I] 50 . Where n ϭ 2, second-order binding is observed, i.e. the degree of polymerization of integrase oligomers is increased by two upon transfer from solution to DNA. In this case, the value [I] 50 is equivalent to ߛK d . Transformed into the familiar Scatchard form, we obtain following the linear expression: To obtain the concentration of free integrase, [I], binding isotherms (constructed in terms of total integrase monomer, [I] T ) were corrected for the stoichiometric number of binding sites (N) (again expressed in monomer units). Since the concentration of bound integrase, [I] B , is given by the fraction () of the total concentration of binding sites (N⅐[D] T ) occupied, we then may write the following: For a series of titrations performed at different concentrations of DNA, at any given value of , plotting [I] T versus ⅐[D] T gives a straight line from which [I] F and N may be determined directly from the intercept and slope of the line, respectively. The binding isotherm is generated by repeating this analysis throughout the range of values, essentially as described previously (41).
Dissociation kinetics were analyzed by the first-order exponential decay expression: where D o is the concentration of bound DNA at time 0, and k is the corresponding dissociation rate constant.

RESULTS
Oligomeric Form of the HIV-1 Integrase-The HIV-1 integrase was expressed and purified as a His-tagged fusion protein (see "Experimental Procedures"). The oligomeric state of this protein was analyzed by gel filtration under conditions of varying ionic strength. In these experiments, a set of calibration standards was used to determine the Stokes radius of the protein, and variations in the volume of the equivalent hydrodynamic sphere were interpreted as reflecting solely changes in the oligomeric state of the protein as the ionic strength was lowered. Whereas the integrase behaves as a monomer at 1 M NaCl (data not shown), at an intermediate salt concentration (0.4 M) both monomeric and dimeric species are observed (Fig. 1). These data are consistent with the monomer-dimer equilibrium previously reported for both HIV-1 (37) and avian myeloblastosis virus (42) integrases on the basis of sedimentation studies conducted under similar salt concentrations. However, at a lower ionic strength more closely approximating that used in our DNA binding assays, we observed that the integrase exists exclusively in a higher order oligomeric state, minimally forming trimers, though more probably tetramers or larger aggregates.
DNA Binding Studies-We first studied the binding of the integrase to oligonucleotide mimics of the viral U5 and U3 LTRs. Nucleoprotein complexes formed between the HIV-1 integrase and the appropriate DNA substrate were irradiated with a single 5-ns pulse of high energy 266-nm UV light, and adduct formation was monitored by SDS-polyacrylamide gel electrophoresis. By this approach, a single, major cross-linked adduct is observed, exhibiting a molecular mass equivalent to that of the HIV-1 integrase (33 kDa) plus the oligonucleotide strand participating in adduct formation (6.8 kDa). A representative example is shown in Fig. 2, where the intensity of the cross-linked complex can be seen to increase as a function of the integrase concentration. The concentration dependence of the binding reaction does not, however, correspond a simple hyperbolic function expected for an independent, single-site binding mechanism. Instead, the sigmoidal nature of the saturation curve indicates the cooperative binding of the integrase to the LTR (Fig. 3A). This is readily seen when viewed in the form of a Scatchard plot (Fig. 3B). The data points display the humped curvature characteristic of cooperatively associating systems (40). For the 21-bp U5 LTR duplex, the positive cooperativity complies with a Hill coefficient of approximately two. Consequently, the data points may be linearized by making a secondorder correction compatible with the cooperative association between two integrase oligomers on the DNA (for details, see "Experimental Procedures").
The yield of cross-linked product is dependent on two factors: the fraction of the total DNA binding sites occupied by the integrase and a quantum yield factor ( c ) defined by the photochemical reactivity of the occluded site (as detailed under "Experimental Procedures"). As shown below, the titration curves obtained in this way correlate well with those obtained by nuclease digestion techniques when measured under identical conditions (see inset to Fig. 6B), thereby confirming that c remains constant over the entire fractional saturation curve. The absolute value of c is however dependent on the photochemical reactivity of the DNA sequence occluded by the bound protein, which varies dependent on its base composition. This is seen readily from the comparison of the maximal crosslinking obtained with single-stranded homopolymers ( Table I).
The photochemical reactivity of the four nucleotide bases may be ordered (relative to poly(dT) ϭ 1) as T (1) Ͼ Ͼ C (0.09) Ͼ A (0.04) Ͼ Ͼ G (0). Thus, information on the binding affinity of the integrase is obtained from the entire fractional saturation curve as described above and not from the absolute value of DNA cross-linking.
Absence of Binding Specificity-In terms of the total concentration of integrase monomers required to reach half-maximal saturation, no preference was observed for the binding of the duplex viral LTR sequences (Table I). The HIV-1 integrase displays a similar apparent affinity for both single-and doublestranded forms of the LTR and for nonspecific single-stranded DNA. Curiously, although no sequence specificity appears to operate in the binding mechanism, some level of preference is evident from the comparison of the titration of homopolymeric single-stranded sequences. Single-stranded polypyrimidine sequences (poly(dT), poly(dC)) are bound with equal efficiency as the LTR, but not polypurine sequences (poly(dA)). The lower affinity seen for poly(dA) when compared to the polypyrimidine sequences has been confirmed in direct competition experiments, where poly(dA) fails to compete out the binding of poly(dC) (data not presented). Similarly, poly(dG), which is essentially refractory to UV cross-linking at 266 nM, competes poorly for the binding of poly(dT) or the U5 LTR (data not shown).
The binding of both LTR and nonspecific sequences is notably weaker in the absence of a divalent cation or when Mn 2ϩ is replaced by Mg 2ϩ . In the absence of a detectable specificity for the cognate LTR sequences, we measured the affinity of binding to poly(dT) as a function of the ionic strength (Fig. 4). Several features of this nonspecific interaction are noteworthy. The binding affinity is markedly sensitive to the concentration of monovalent cations. Consistent with the polyelectrolyte theory (43), log K obs (the observed association constant, in this case reported in terms of total monomer) decreases linearly with log [NaCl] (in the range 0.05-0.2 M NaCl). For simple equilibria, this approach may be used to extract information on the number of DNA-phosphate:Na ϩ ion pairs extruded by the binding of a protein ligand to DNA. However, it should be kept in mind that other factors, such as the influence of NaCl on the oligo-  Fig. 2. The sigmoidal curve, shown in semi-log form (panel A), represents the best fit of the data by non-linear least squares procedures in terms of the total HIV-1 integrase concentration and corresponds to a Hill coefficient of 1.90 Ϯ 0.2 (see "Experimental Procedures"). In panel B, the data are transformed, in terms of fractional saturation, to Scatchard form for a first-order (top) and second-order (bottom) reaction. In this case, the data points are corrected for the bound protein assuming a density of 10 monomers per LTR at saturation (as defined in Fig. 6). The Hill coefficient is 1.94 after transformation. meric state of the integrase, may also affect the apparent affinity. Nonetheless, the data suggest that binding to singlestranded DNA may involve an important electrostatic contribution. Such "salting-out" effects on binding were also observed at the higher concentrations of Mg 2ϩ , although an initial stimulation of binding was observed with this cation. In contrast, no significant difference in K obs was noted when the concentration of MnCl 2 was varied between 0.1 and 10 mM. This demonstrates that the Mn 2ϩ cation may stimulate binding at considerably lower concentrations than have been reported previously for the optimum enzymatic processing of the LTR substrates (27).
Stable Complexes Resist Attack by Nucleases-Binding of the LTR by HIV-1 integrase results in the formation of a stable complex that resists challenge by DNA or heparin (26,29,44). To investigate the assembly of such complexes, a novel approach was developed whereby their formation was probed by way of exposure to the 3Ј-5Ј-specific exonuclease, snake venom phosphodiesterase I. The LTR is rapidly degraded when incubated directly with the nuclease; greater than 95% of the DNA is attacked (shortened by one nucleotide or more) within approximately 1-2 s. Complete reduction to single nucleotides occurs within 30 s (data not shown). The degradation is, however, significantly reduced upon the pre-incubation of the LTR with the integrase (Fig. 5, left panel, lane 1). Competitor DNA completely eliminates the protection of the LTR if added before (lane 2), but not after (lane 3), the pre-binding step. The protection of the LTR by the integrase is observed in an all-or-none fashion, irrespective of the strand bearing the radiolabel. The absence of intermediate length products demonstrates that both ends of the DNA molecule are bound coincidentally in the stable complex. Furthermore, the retention of the Ϫ2 base cleavage product confirms that the substrate 3Ј-end remains tightly bound and protected from the nuclease following processing. Indeed, the respective dissociation kinetics for the U5 LTR when present in blunt-ended or pre-processed form (lacking the two terminal bases) are indistinguishable (k diss ϭ 1.5 ϫ 10 Ϫ5 s Ϫ1 ). Stable binding does not appear to be sequence specific. Complexes with oligonucleotide substrates bearing no obvious sequence similarity to the LTR exhibit equivalent stability (data not shown). Notably, however, neither strand of the LTR may be protected when present in single-stranded form (Fig. 5, right panel). This contributes clear evidence that the nonspecific binding of HIV-1 integrase to single-and doublestranded DNA involves different binding conformations that may be distinguished on the basis of their stability in dissociation experiments.
Stoichiometry of the Stable Nucleoprotein Complex-In accordance with the UV cross-linking data (above), the binding reaction leading to nuclease-resistant complexes also displays a second-order dependence on the total integrase concentration (Fig. 6). To obtain a quantitative analysis of the binding density of the integrase, titrations were performed with several different concentrations of the DNA substrate, thereby allowing the bound and free forms of the integrase to be discriminated on the basis of the conservation of mass (41). An example of this Ͻ0.01 a Oligonucleotides are shown in single-stranded form or hybridized to their complements except for disU5, which is a single oligonucleotide capable of forming a branched (dumbell) structure (24).
b The concentration of integrase in total monomer required to achieve half-maximal saturation. c The maximal percent of DNA cross-linked, equivalent to C , a quantum yield factor dictated by the sequence composition of the oligonucleotide (see under "Experimental Procedures"). type of analysis is shown in Fig. 6C. The concentration of integrase required to achieve half-maximal saturation (in terms of the fraction of DNA protected) indicates that the bound complexes are constructed from approximately 10 monomers per 21-bp DNA duplex. The free enzyme concentration is obtained by extrapolation of the data to the intercept with the vertical axis, where the concentration of bound integrase is effectively zero. The broken line in Fig. 6B traces the theoretical binding isotherm, in terms of the free integrase concentration, obtained from the systematic repetition of this type of analysis throughout the fractional saturation curve. The solid lines in Fig. 6B also represent theoretical second-order binding curves. These were calculated from the binding isotherm after transformation to account for a binding density, at saturation, of 10.5 monomers per LTR. In this way, good accordance between data and theory is seen, demonstrating that the binding density obtained is a fair approximation of that existing over the entire binding curve.
Divalent Cation Requirement-The formation of the stable complex does not exhibit an absolute requirement for the presence of a divalent cation. Nuclease-resistant complexes appear, albeit at low efficiency, when the integrase is incubated with the U5 LTR in the absence of the cation (Fig. 7). The protection of the LTR is enhanced considerably in the presence of Mn 2ϩ but only poorly with Mg 2ϩ cations. These results are entirely consistent with the influence of the respective divalent cations on the apparent DNA binding affinity reported in Table I. A sigmoidal response, in terms of activity versus cofactor concentration, is predicted to arise from a cofactor-induced oligomerization of a protein where the oligomeric form interacts preferentially with the substrate. In the case of the integrase, Mn 2ϩ clearly enhances binding in a non-hyperbolic fashion. The Hill coefficient (n h ) measured for the Mn 2ϩ -induced transition observed in Fig. 7 is 4.2 Ϯ 1.0. A similar value (n h ϭ 5.0 Ϯ 1.2) was obtained after the proteolytic removal of the His-tag sequence from the recombinant protein, confirming that this property is inherent to the integrase. Consequently, the kinetics of 3Ј-end processing are also markedly non-linear with respect to the Mn 2ϩ concentration (data not shown). These data are thus consistent with the cooperative binding mechanism and indicate that the metal ion promotes the proteinprotein interactions required for the stable binding of the DNA substrate.
A further remarkable feature of the transition is the concentration range over which it is effected. The apparent binding constant (K app ) for Mn 2ϩ estimated from several such titrations, for both His-tagged and factor Xa cleaved proteins, is 64 Ϯ 10 M. Whereas the optimum in vitro 3Ј-end processing and disintegration activities of the integrase are obtained at high (non-physiological) concentrations of Mn 2ϩ , comparatively poor activity is exhibited with Mg 2ϩ (27,37,45). To further investigate the role of the divalent cation in the formation of the stable complex, the respective dissociation and 3Ј-end processing kinetics of nuclease-challenged complexes were compared (Fig. 8). Complexes formed with Mn 2ϩ are considerably more stable than the corresponding complexes formed with Mg 2ϩ , exhibiting half-lives of approximately 13 and 1 h, respectively. However, the first-order rate constant obtained for the 3Ј-end processing reaction is similar in the presence of either cation. By eliminating the uncomplexed DNA, we follow only the enzymatic activity of the stable complex. The rate constant obtained then gives a direct measurement of k cat , which, under these conditions, was 0.046 Ϯ 0.011 min Ϫ1 and 0.069 Ϯ 0.008 min Ϫ1 for Mn 2ϩ and Mg 2ϩ , respectively. Such low rates are consistent with the delayed appearance of processed 3Ј-termini in virus-infected cells, which accumulates over a period of several hours (1), and is probably an intrinsic function of the catalytic properties of the integrase. Nonetheless, we may conclude that the overall higher efficiency of 3Ј-end processing in Mn 2ϩ observed in vitro may be explained simply by the augmentation in the stability of the interaction with the substrate and not to an inherent preference for this cation for catalytic function.

DISCUSSION
In this report, we have described the development of novel procedures to probe the interaction between HIV-1 integrase and DNA in vitro as the preliminary step in a detailed study of the structural architecture of these nucleoprotein complexes. Whereas the first approach, laser UV cross-linking, detects binding via the intimate protein-DNA contacts formed within an occluded photoreactive site, for nonspecific binding it is indiscriminate as to the position of that site on the DNA molecule. On the other hand, protection of the DNA from hydrolysis by exonucleases detects principally the stable contacts formed with the ends of the DNA substrate. The fact that identical binding isotherms are generated by these two different techniques, combined with the absence of intermediate  1 and 3) or presence (lane 2) of competitor DNA. In lane 3, competitor DNA was added after the pre-incubation period and 5 min before the nuclease. In lane 4, the integrase and nuclease were pre-mixed before addition of the U5 LTR. The positions of the full-length (21-mer), 3Ј-processed (19-mer) strands and free nucleotides (nt) are shown (the appearance of a Ϫ3 base product with the U5 duplex is due to the 3Ј-end processing of a quantity of Ϫ1 base substrate carried over from the oligonucleotide synthesis after gel purification). Corresponding activities with the U5 LTR in single-stranded form are shown in the right panel. Lanes 5-8 were with the U5a (cleavage) strand, and lanes 9 -12 were with the complementary U5b strand (see Table I). Lanes 5 and 9, incubations performed with integrase alone. Lanes 6 -8, and 10 -12, nuclease was added after pre-incubation of the integrase with the DNA. Competitor DNA was added either after (lanes 6 and 10) or during (lanes 7 and 11) the pre-incubation step or omitted entirely (lanes 8 and 12).
length nuclease digestion products, strongly implies that the entire DNA molecule, including both ends, is occluded upon the formation of the stable nucleoprotein complex. Since non-occluded regions and unbound ends appear to be absent (both of which, if present, could potentially have introduced anti-cooperative effects (40)), the binding isotherm is expected to be influenced only by the positive cooperativity involved in this mechanism. In this case, the corresponding Hill coefficient should be representative of the number of oligomers participating in the assembly of the stable complex (see "Experimental Procedures"). For both techniques, the second-order binding observed with the 21-bp LTR is consistent with a mechanism involving the cooperative dimerization of a high-order oligomeric form of the protein. From a density of 10.5 Ϯ 0.8 monomers per LTR and a Hill coefficient of 2, the oligomeric form of the integrase in solution is estimated to be pentameric, though, within the limits of reasonable experimental error, the average degree of oligomerization (M) in solution may be extended to incorporate tetrameric or hexameric forms (4 Յ M Յ 6). Clearly, the resultant high-order nucleoprotein complex should exhibit a substantial molecular mass (Ͼ300 kDa). This may, in part, provide an explanation for the sedimentation properties of such complexes (13) and may indicate the reason why such complexes are not particularly suited to gel retardation studies.
We have evaluated the cooperativity for the binding of short oligonucleotide duplexes only. In this limiting case, the observed cooperativity may be restricted, by reason of the number of bases occluded by a single oligomer, to a single cooperative (apparent dimerization) event. Recent experiments have enabled us to observe the complexes formed with larger DNA fragments directly by electron microscopy. In the presence of Mn 2ϩ , large clusters of bound integrase molecules, extending inwards from the DNA ends, have been detected, indicating that the associated cooperativity parameter may be quite large. 2 These findings appear to comply with the extended, nucleosome-like DNase I footprints observed for the interaction between the avian myeloblastosis virus integrase on plasmidborne LTRs (46). Accordingly, the cooperativity parameters will require further investigation with longer DNA substrates. We may, however, infer from these data, that in the absence of other cellular factors, the integrase oligomer binds contiguously to naked DNA rather than to distinctly separated sites. substrates thus presents itself as a useful model system with which to examine binding under conditions of limited cooperativity.
In agreement with several previous reports (6,11,(13)(14)(15)(16)(17), no sequence specificity was observed in the binding of the viral LTR nor was a preference apparent for the double-stranded substrate in UV cross-linking titrations. Stable complexes are, however, not formed with single-stranded DNA, suggesting the underlying mechanism of binding may be different for these two forms of DNA. Our recent data suggest that the binding density of integrase for single-stranded DNA (poly(dT)) is considerably lower than that observed for the U5 LTR duplex. 3 We expect that the differential dissociation rates obtained with such complexes may reflect this observation. We are currently investigating this in more detail.
We have observed that Mn 2ϩ can stimulate binding at concentrations between one and two orders of magnitude lower than that previously reported as the optimum concentration for 3Ј-processing for both the HIV-1 (27) or Moloney murine leukemia virus (45) integrases. An explanation for this disparity could be in the existence of two distinct divalent cation binding sites on the integrase oligomer: one involved in mediating protein-protein interactions, and hence stimulating cooperative binding, and another incorporated at the catalytic site. Indeed, a region distinct from the catalytic site, the HHCC finger domain, has recently been implicated in the oligomerization of the HIV-1 integrase in the presence of divalent cations, a property further shown to be essential for stable complex formation (26). Such interactions probably underlie the cooperativity of DNA binding, much as the coordination of Zn 2ϩ by the HX 12 CX 9 CX 2 C motif (47) of the T4 bacteriophage gene 32 protein (gp32) has been inferred to stimulate cooperative binding to single-stranded DNA via enhanced protein-protein interactions (48). The presence of Zn 2ϩ is not necessary to obtain cooperative binding of gp32 but stimulates it approximately 30-fold (48). By analogy, the coordination of a metal ion by the integrase HHCC motif may stabilize a particular conformation of this subdomain and thereby enhance a property already inherent in the integrase to self-associate. Although the precise Mn 2ϩ interaction site remains to be resolved, the general assumption would appear to be correct. Stable complexes form in the absence of the divalent cation but only efficiently at much higher protein concentrations. 3 Thus, the cation appears to favor the polymerization process by a mechanism that lowers the activation energy barrier to this association. A similar observation has been made for the self-association equilibrium of the RecA protein, which is stimulated in the presence of Mg 2ϩ (49). The capacity of such proteins to associate in the absence of DNA appears to be a recurring theme for proteins that are known to form direct and functional interactions in the context of a highly structured DNA-protein complex.
We have demonstrated that the ternary nucleoprotein complex formed with Mn 2ϩ is considerably more stable than the corresponding complex formed with Mg 2ϩ . The absolute rate of 3Ј-end processing exhibited by the stable complex is, however, similar with either cation. This allows us to deduce that the marked preference exhibited in vitro for Mn 2ϩ is simply a consequence of the enhanced stability of binding to the LTR. Furthermore, it was reported (after submission of this manuscript) that an increase in the efficiency of the 3Ј-end process- Integrase (350 nM) was pre-incubated with the U5 LTR (12.5 nM) for 5 min prior to incubation with phosphodiesterase for 10 min at 37°C. The plot shows the consequence of altering the concentration of MnCl 2 (squares) and MnCl 2 (triangles) on the extent of DNA protected.
FIG. 8. Comparison of the kinetics of dissociation and 3-end processing exhibited by stable complexes formed in the presence of Mn 2ϩ or Mg 2ϩ . Integrase (390 nM) was pre-incubated with the U5 LTR for 3 min at 37°C prior to addition of phosphodiesterase I. Incubations were stopped at appropriate time intervals and analyzed by gel electrophoresis to discriminate the percentage of the original substrate remaining and the extent of 3Ј-end processing of the stable complexes (as detailed under "Experimental Procedures"). Left panel, kinetics of dissociation (circles) and 3Ј-end processing (triangles) in the presence of 5 mM MnCl 2 . Right panel, kinetics of dissociation (circles) and 3Ј-end processing (triangles) in the presence of 5 mM MgCl 2 . The time reported for dissociation is with respect to addition of the nuclease. For 3Ј-end processing, this refers to total incubation time. Data points are fitted by non-linear least squares analysis to the appropriate first-order expression.
ing of the LTR substrate in Mg 2ϩ can be achieved either by altering the solution conditions (notably to include PEG) (50) or by increasing the length of the DNA substrate (51). In both cases, these effects can be explained simply by a concommitant stimulation in the binding interaction with the DNA substrate and add further evidence to support our conclusion that the bias for Mn 2ϩ in standard in vitro reactions operates at the level of DNA binding and not catalysis. The requirement for Mn 2ϩ is not seen with the pre-assembled viral nucleoprotein complex purified from infected cells, which undergoes efficient integration in vitro in the presence of Mg 2ϩ (52). Clearly, since the preference exhibited for Mn 2ϩ in vitro is to stimulate efficient binding of the recombinant protein to the DNA substrate, the necessity for this cation has already been circumvented when using the preformed integration machinery. Other cellular components, potentially host and/or viral accessory proteins, may be required to regulate the assembly of the authentic nucleoprotein complex. In this respect, the presence of several viral proteins other than integrase has been noted in the pre-integration complexes (5). In the absence of such factors, the physiological role of the metal ion-induced interaction between integrase oligomers remains unclear. It is possible that the integrase binds to the viral DNA ends as a structured and highly stable aggregate. It also appears likely that some interaction should take place between integrase molecules bound at the extreme ends of the viral genome to allow the coordinated integration of the viral termini at a discrete target site in the host genome. This is accomplished in the analogous transposition mechanism of bacteriophage Mu by coordination of distant MuA binding sites into a tetrameric MuA synaptic complex (53). By comparison, a direct interaction between the distally bound integrase complexes may bring the LTR termini together to form the legitimate strand-transfer complex. Indeed, our recent observations 3 and those of Mazumder et al. (33) suggest that the assembly of complexes is not restricted to lateral, cooperative interactions, but may also involve intermolecular interactions between bound complexes. Undoubtedly, a greater understanding of the process of retroviral integration will necessitate a more detailed exploration of the structural architecture and cofactor requirements of the functional integrase nucleoprotein complexes. We expect that the approaches outlined in this report will greatly assist these studies.