Architecture of recombination intermediates visualized by in-gel FRET of λ integrase–Holliday junction–arm DNA complexes

Results

Experimental Strategy. The position of the two arm sites on P′1,2 arm DNA relative to the four core sites on the HJ in Int ternary complexes can be determined by measuring the distances between arm sites and core sites by using a FRET-based method. To assemble unique one-arm ternary complexes (i.e., complexes in which the arm DNA is bound to a unique nonpartner pair of Ints), we took advantage of bispecific HJs containing a mixture of Int binding sites with different specificities (17). In the HJ substrate diagramed in Fig. 2, one pair of nonpartner core sites is specific for λ Int (λ sites), and the other is specific for the closely related HK022 coliphage Int (HK sites).

HK022 Int is very similar to λ Int in structure and function, but it recognizes different core-type sites, which is the only known functional difference between these two proteins (28). We used a previously characterized λ-HK022 chimeric Int with the core recognition specificity of wild-type HK022 (29, 30). The N-terminal domain and adjoining linker region of the chimera are identical in sequence to λ Int, and the two proteins have been shown to cooperate well during resolution of bispecific HJs (17, 29).

An arm-binding defective mutant of λ Int (N^–) that is altered in the extreme N terminus (31) was used in conjunction with the arm-binding proficient λ-HK022 chimeric Int protein (N⁺). Both N⁺ and N^– proteins are catalytically inactive because of a Tyr342Phe (IntF) mutation that eliminates the Tyr nucleophile. P′1,2 arm DNA, the N⁺ and N^– IntF variants, and the bispecific HJs assemble into one-arm ternary complexes in which the arm DNA can bind only to the N⁺ nonpartner pair (Fig. 2) (17). Neither protein alone can form tetrameric (or trimeric) complexes on the hybrid HJs and in the presence of arms neither protein alone can form ternary complexes (data not shown; ref. 17). The arm binding defect of N^– IntF (31) was confirmed by its failure to form ternary complexes with an HJ containing all four λ sites (data not shown).

The substrates used in this study are immobilized synthetic HJs in which the crossover position and predominant isomer were shown to bias Int-mediated resolution exclusively in one direction (32, 33). In all of the figures (e.g., see Fig. 2), the preferentially cleaved and exchanged DNA strands are shown as continuous lines and referred to as the exchanging (Ex) strands. The other pair of strands are shown as dashed lines and referred to as the nonexchanging (NEx) strands. The core sites that bind the “active” pair of partner Ints (positioned to cleave the Ex strands) are designated λ (Ex) and HK (Ex), whereas the core sites that bind the “inactive” pair of partner Ints (positioned to cleave the other strands) are designated λ (NEx) and HK (NEx). Two types of HJs were used (Fig. 2). In HJ1, the HK sites (bound by the N⁺ proteins) are separated by the angle formed by an Ex strand. In HJ2, the HK sites are separated by the angle formed by a NEx strand.

Complex Assembly and FRET Measurements. We developed an in-gel FRET method that enabled us to map the position of the P′1,2 DNA in the ternary complexes described above by measuring 15 distances between six points within the complex (the two arm sites and the four core-sites; Fig. 2). The efficiencies of energy transfer between donor and acceptor dyes at various positions were determined from the extent of donor fluorescence quenching in complexes containing the acceptor compared with donor fluorescence in complexes without acceptor.

We measured FRET in both HJ1 and HJ2 ternary complexes with the dye attachment sites in the two ternary complexes as diagramed in Fig. 2 Lower.

To measure the distances between each arm and each core site, a radioactively labeled HJ with the donor attached next to one of the core sites was mixed with N^– and N⁺ IntF proteins and various amounts of unlabeled P′1,2 or P′1,2 DNAs carrying the acceptor next to P′1 or P′2 (P′1(A)P′2 or P′1P′2(A), respectively). After native PAGE, the fluorescence intensities of the donor in the complexes with and without acceptor were obtained from laser scans of the gel. The total amount of complex formed in each lane was determined from the amount of radioactive label in each complex (quantitated from a PhosphorImager scan of the same gel) (Fig. 3A).

The efficiency of energy transfer, E, is calculated from the slope (1 – rE) of the line plotted in Fig. 3B, where r is the efficiency of labeling of the DNA substrate with the acceptor dye (see Materials and Methods). The distance R between donor and acceptor is calculated from the measured E value by using the Förster equation for FRET: where R ₀ is the Förster radius determined for each donor–acceptor pair (see Materials and Methods, Supporting Text, and Table 5). Calculations using an average R ₀ produced similar structures as those described below (data not shown).

The table below the plot in Fig. 3B shows the measured E and calculated R values for the λ (NEx)/P′1 and λ (NEx)/P′2 donor–acceptor pairs in the HJ1 complex. A slope of 1.1 for λ (NEx)/P′2 indicates that no FRET has occurred (E = 0), suggesting that the donor and acceptor are >100 Å apart (i.e., more than twice their Förster radius). However, E for the λ (NEx)/P′1 pair was 27%, and therefore, the distance between these two points in the HJ1 complex is ≈60 Å.

To determine the distance between core sites, the acceptor dyes were attached to the HJ next to core sites and donor fluorescence in donor-only HJs was compared with donor fluorescence in donor-plus-acceptor doubly labeled HJs (the P′1,2 DNAs were unlabeled, and both HJs were radioactively labeled). To estimate potential arm-DNA bending in the ternary complexes, the distance between the two-arm sites was determined from ternary complexes assembled with a donor-only labeled or a donor-plus-acceptor doubly labeled P′1,2 DNA (also radioactively labeled) and an HJ without any fluorescent tag.

All of the measured energy transfer efficiencies and calculated distances between each donor–acceptor position in HJ1 and HJ2 complexes are summarized in Table 1.

Coordinate Determination and Structure Modeling. We used a metric matrix distance-geometry computer program, standardly used for NMR structure determination (34), with minor modifications, to calculate the spatial coordinates of six dye positions by using the 15 distance ranges measured by FRET and summarized in Table 1 (see Materials and Methods). This method provides completely unbiased searching of all possible topologies of the six points fulfilling the experimental observations. Note that FRET methods for structure mapping cannot discriminate between the structures shown in Fig. 4 and their mirror images. Indeed, coordinate calculations generate nearly equal numbers of both mirror images, only one of which is represented in our figures (see Materials and Methods). The distance between HK (NEx) and P′1 in the HJ1 complex did not fit with the other distance constraints and had to be discarded. The diagrams in Fig. 4A show the expected locations and alternative directions of arm DNAs in the HJ1 and HJ2 complexes.

The fact that arm DNA in the ternary complexes deviates only slightly from a B-DNA structure is shown in Fig. 4C, in which the P′1 and P′2 data points (blue) from Fig. 4B are superimposed on the fluorophore attachment points (dark gray) in a model of a B–DNA duplex representing arm DNA. However, superposition of fluorophore attachment points (light gray in Fig. 4C) on an HJ model structure derived from a Cre/HJ crystal structure (27) and the experimental HJ points (green in B and C) from the ternary complexes suggests that the HJ in Int ternary complexes is less planar than the Cre/HJ model and, instead, seems to adopt an X-shape with two pairs of adjacent arms lying in two intersecting planes.

The results presented above demonstrate that in-gel FRET analysis could be used to derive a coherent self-consistent 3D map of the Int–HJ–arm DNA ternary complexes. Because 15 FRET distance ranges (all of the possible distances between six points) were used for coordinate calculation, the positions of the six fluorophores in the ternary complexes could be determined with good precision (see Fig. 4B and the standard deviations from D values in Table 1) despite the wide distance range of individual FRET distance measurements (see R in Table 1). The map yielded a defined orientation of arm sites on P′1,2 DNA relative to core-sites on the HJ (clockwise in the view shown in Fig. 4). In addition to their directionality, the map surprisingly also revealed that the arm sites are not stacked directly above the HK core sites, as expected. The arm DNA is instead rotated ≈45° around the z axis relative to its predicted position in Fig. 4A (Fig. 4D). In this configuration, the P′1 site is located close to the λ (NEx) or λ (Ex) sites in the HJ1 and HJ2 complexes, respectively. The positioning of an arm site above a λ site occupied by a N^– (arm-binding-defective) IntF suggests that the N-terminal domains in the Int tetramer are not stacked directly above their respective C-terminal domains (i.e., above the C-terminal domains to which they are physically attached), but rather, they are extended, or “flipped out” and are stacking above the neighboring Int C-terminal domain, as shown in the diagrams in Fig. 4D.

Resolution of HJ1 and HJ2. In the FRET geometric maps derived above, the P′1,2 DNA in HJ1 spans the angle formed by a NEx strand, whereas in HJ2, the P′1,2 DNA spans the angle formed by an Ex strand (Fig. 4D). Is the difference between HJ1 and HJ2 functionally significant? This question is important because it impacts significantly on understanding the recombinogenic complexes (see Discussion).

In the absence of arm DNA, the HJ substrates used in this study are resolved in only one direction: by the cleavage and exchange of the strands shown as continuous lines in the diagrams and named the Ex strands. To detect all of the potential resolution products, the HJ substrates used in Fig. 5 were constructed with each of the four branches being a different length. As a result each of the possible resolution products would be a different size and would migrate to a unique position during PAGE.

Fig. 5 shows the results of a time course of HJ resolution mediated by a mixture of N^– Int (bound to λ sites) and N⁺ Int (bound to HK sites). The experiment was performed at very low protein concentrations to ensure protein binding specificity, although at the expense of resolution rates. In the absence of arm DNA, both HJ1 and HJ2 are resolved at comparable rates, and no products resulting from the cleavage and exchange of NEx strands were detected with either HJ (Fig. 5 and data not shown). Although arm DNA is known to stimulate HJ resolution (17), we sought to determine whether HJ1 and HJ2 differ in their response to arm DNA. Although HJ2 is stimulated ≈2-fold by arm DNA, HJ1 does not respond. The observed 2-fold stimulation of HJ2 resolution by P′1,2 DNA is likely to be an underestimate because, at these low protein and arm DNA concentrations, most of the HJ resolution products probably result from Int–HJ complexes that do not have any arm DNA. Under all of the examined conditions, the only detected resolution products are still a result of cleaving and exchanging the Ex strands. The difference in HJ resolution between HJ2 and HJ1 complexes strongly suggests that the arm-DNA configuration in the former is the one similar to the structure of the full HJ recombination intermediate.

Discussion

In-Gel FRET Structural Mapping. The TIF mapping of the spatial relationships between components in higher-order complexes is particularly well suited for heterogeneous mixtures in which several different complexes are in equilibrium and the ability to focus on a single complex is critical. It affords a straightforward and facile approach to determining the global organization of large higher-order complexes, and its applicability should be enhanced further by the site-specific incorporation of fluors into individual proteins (35). The power and resolution of this analysis derives from distance-geometry calculations (34), which converge to a unique set of coordinates with very small errors (see D in Table 1) despite the large range in the measurement of individual FRET distances (R in Table 1). In our coordinate calculation, each single pairwise distance between six dye positions is constrained by 14 other pairwise distances, thereby generating precise coordinates for these six points. Somewhat different approaches in the analysis of structural maps derived from FRET measurements have been used in other systems, such as distance-constraint docking of the σ⁷⁰ subunit to RNA polymerase and the global positioning of a maltose-binding protein–quantum dot conjugate (35, 36).

Structural and Functional Features of HJ1 and HJ2 Complexes. Two important structural insights were derived directly from the structures of the HJ1 and HJ2 ternary complexes shown in Fig. 4. First, the direct repeat P′1 and P′2 arm sites adopt a clockwise direction relative to core sites as shown in Fig. 4. Second, arm sites are not stacked directly above the HK core sites as expected. To appreciate this second point, it should be noted that the model shown in Fig. 4D fits the observed data 5- to 10-fold better than it fits a model with direct stacking of arm and core sites (Fig. 4A); i.e., distances between the surfaces of the spheres representing the sites of dye attachment (dark gray spheres in Fig. 4C) to arm DNA and the surfaces of the spheres representing the positions of fluorophores in space as determined by FRET analysis (blue spheres in Fig. 4C) would have to increase 5- to 10-fold to accommodate the direct stacking model (see Table 6, which is published as supporting information on the PNAS web site). Consequently, the only simple interpretation for the P′1 arm-DNA site being positioned over Int protomers with a defective N-terminal domain is that N-terminal domains of Int are extended, or flipped out, toward the neighboring Int subunit in a counterclockwise direction in the view shown in Fig. 4D. The TIF mapping and the models derived from it are robust and convincing; however, an additional validation in support of this approach is the excellent agreement between the results reported here and the x-ray crystal structures of the ternary Int–HJ–arm DNA complexes (44).

Although the structural features described above are common to both the HJ1 and HJ2 ternary complexes, the two complexes differ by 90° in the orientation of the P′1,2 DNA relative to the HJ overlap region; in HJ1, P′1,2 DNA crosses the angle formed by an NEx strand, whereas in HJ2, it crosses the angle formed by an Ex strand. The results shown in Fig. 5 indicate that P′1,2 DNA increases the efficiency of HJ2 resolution but has no effect on HJ1 resolution. There are no large structural differences (i.e., >10- to 15-Å resolution of FRET) between the approximately square planar HJ conformations in HJ1 and HJ2 (Fig. 4 and Table 1), and both ternary complexes form with equal efficiency and have equal stability (data not shown). Therefore, we propose that when arm DNA is bound across the Ex strand HJ angle, as in HJ2, it induces conformational changes that stimulate the active Int partner pair, either directly or by means of subtle changes in the HJ conformation. However, this stimulating effect is not sufficient to overcome the constraints and Ex strand selectivity imposed by the sequence of the overlap region because the NEx strand in HJ1 is not cleaved, even in the presence of arm DNA.

Because arm DNA stimulates only HJ2, and not HJ1, resolution, we propose that the arm DNAs cross the HJ angles defined by the Ex DNA strands in the full HJ recombination intermediates (the Ex strands in the full recombination reactions are referred to as the bottom strands) (see Fig. 6 and text below).

Isolated P′1,2 arm DNA is bent ≈45–55° when bound by adjacent Int protomers, and it is also bent within the ternary complexes (11, 17, 37). This study indicates that, within the ternary complexes, this bend is probably <40° (the limit detectable by FRET) because we did not observe a decrease in the end-to-end distance of the arm DNA in the ternary complex compared with unbound arm DNA (Table 1). For slightly curved or straight arm DNAs, the interactions between the N-terminal domains in the tetramer are only 2-fold symmetric; i.e., the interactions between the two N-terminal domains bound to the same arm-DNA duplex differ from the interactions between two adjacent N-terminal domains bound to different arm DNAs. This 2-fold symmetry explains why arm DNA stimulates resolution of only one partner pair within the Int HJ tetramer. Furthermore, this bias is influenced by nonequivalent interactions between the N-terminal domain of one Int and the linker region of an adjacent protomer within the tetrameric Int—HJ complex (38).

Models for the Full HJ Recombination Intermediates. The simplifying assumptions required for the interpretation of genetic data based on synthetic lethality (14) are now determined to be incompatible with the recent insights regarding flipped-out N-terminal domains (this work) and the nonequivalent interactions between N-terminal domains and linker regions of adjacent Int protomers (38).

The models of the integrative and excisive recombination intermediates shown in Fig. 6 incorporate the findings reported here, such as the clockwise directionality of the arm sites, the counterclockwise flipping-out of the N-terminal domains, and the arm DNA crossing of the Ex strands, as well as older findings, such as the differential accessory-site usage in excision versus integration (16, 39, 40) and the antiparallel orientation of arm sites in the P and P′ arms (17). Because only three arm sites are used either in excision or integration (14, 16), the models are drawn with one of the N-terminal Int domains bound nonspecifically to DNA sequences adjacent either to the P2 or the P1 sites for excision or integration, respectively. Consistent with this feature of the models is the finding that Int can form ternary two-arm HJ complexes in which one arm DNA contains two arm sites, whereas the other contains one arm site plus a potential nonspecific binding site (17).

The two versions of the integrative and excisive models shown in Fig. 6 differ only in the path of the P arm DNA between the C and H1 sites in the integrative complex and between the C and Xis/Fis sites in the excisive complex. This short segment of DNA passes either under or over the HJ plane in Left and Right, respectively. The versions in Left are derived by superimposing the IHF-H′ cocrystal structure of Rice et al. (41) onto the other two IHF binding sites H1 and H2. The versions in Right are derived by extension of the relative pairwise bending patterns of IHF bound to its cognate sites in the absence of Int protein (25). We do not yet understand how either version of the models relates in a simple way to the determinations of global topology and tangle analysis (42, 43).

Although choosing between the two versions of each model will require further biochemical and structural analysis on the entire 450-kDa recombination complex, these data strongly support the basic relationships between individual arm and core sites as mediated by the flipped-out N-terminal domains. In the integrative complex, the N- and C-terminal domains of the same Int protomer form bridges P1–C′, P′2–C, P′3–B′, and nonspecific DNA–B (Fig. 6 Upper). In the excisive complex, the bridges are P′1–B, P′2–B′, P2–C, and nonspecific DNA–C′ (Fig. 6 Lower).

Implications for the Mechanism of Coordination of Int Activity During Recombination. Based on the position of arm-DNA segments relative to the HJ (Fig. 4) and the specificity for the HJ2 in arm-DNA stimulation of resolution (Fig. 5), we propose that arm-DNA segments positioned as shown in Fig. 6 stimulate HJ resolution by means of the cleavage of the bottom strands. Then, how is the Int partner pair cleaving the top strands activated in the first part of the reaction to generate the HJ, and how does Int activity switch from top to bottom strand cleavage (Fig. 1 A) (6, 7)? A gross movement of arm-DNA segments during recombination is unlikely (because of structural constraints imposed by the accessory DNA bending proteins at three specific loci within the complex), and we prefer to invoke a more subtle change in arm-DNA conformation.

According to this view, arm-DNA bending within the synaptic complex favors formation of protein–protein interactions that preferentially activate the Int partner pair bound to C and B sites. After top-strand cleavage and/or HJ formation, the arm DNAs straighten and induce a switch in the conformation of the Int tetramer that activates the bottom strand cleaving partner pair bound to C′ and B′.

Although this mechanism is highly speculative, it is based on a considerable amount of evidence that Int-dependent bending of arm-DNA is context-dependent. On isolated DNA fragments, Int binding at P′1,2 induces a bend of ≈40° (37). In ternary complexes containing Int, arm DNA, and a single-duplex core DNA (similar to the complexes described in ref. 19), the P′1,2 DNA is bent ≈90° (M.R.-L., unpublished data). In ternary complexes with HJ substrates, the P′1,2 DNA appears to have a bend (17), which is <40° (this study). The molecular mechanisms associated with this potential switch from bent to unbent arm DNA are unclear, but studies on a precleaved synaptic complex are likely to be informative. It will be interesting to see how these mechanisms relate to the many heterobivalent cousins of λ Int.