LRET experiments employ a luminescent lanthanide donor and fluorophore acceptor pair with an appropriate Förster radius (R₀) (Zoghbi and Altenberg, 2018). Due to the long lifetime of the luminescence signal, LRET has several advantages over FRET. First, the lanthanide luminescence excited state lifetime is milliseconds, whereas the fluorescent acceptor lifetime is nanoseconds. This difference allows LRET to be measured after background fluorescence has decayed, eliminating artifacts from direct acceptor excitation/emission and scattering by the sample. Second, LRET measurements are insensitive to incomplete labeling because millisecond timescale donor-sensitized acceptor signal only occurs when a donor and fluorophore are in proximity. LRET probes are introduced into a protein most easily through a thiol-maleimide reaction with a unique cysteine. For Pf Rad50^NBD, where the coiled-coil domains are truncated, single cysteines were introduced into the naturally cysteine-less construct, and MR^NBD activity was tested to ensure an active complex (Figure 1—figure supplement 1). Mutations were made primarily in loop regions to minimize disruptions to the protein fold. In all, six separate single-cysteine mutations were made throughout Rad50^NBD (Figure 1B). These cysteines were subsequently labeled with thiol-reactive LRET donor or acceptor molecules. To make the MR^NBD complex for LRET experiments, equimolar amounts of donor-labeled (Tb³⁺-chelate) Rad50^NBD and acceptor-labeled (Bodipy FL or Cy3) Rad50^NBD were mixed with twice the molar ratio of Mre11 so that 50% of the resulting M₂R^NBD₂ complexes had one donor and one acceptor fluorophore (on separate Rad50^NBD protomers). Not only were identical cysteines mixed within a complex (e.g., Tb³⁺-S13C and Bodipy-S13C), but complexes were also made where cysteine mutants were mixed with other cysteine mutants (e.g., Tb³⁺-S13C and Bodipy-L51C). Additional distance measurements were obtained for a given LRET pair by changing the identity of the acceptor as the Förster radius for Tb³⁺ and Cy3 (61.2 Å) is longer than that of Tb³⁺ and Bodipy FL (44.9 Å). Finally, donor- and acceptor-labeled cysteines within mixed pairs were swapped (e.g., Tb³⁺-S13C and Bodipy-L51C versus Tb³⁺-L51C and Bodipy-S13C) for added confidence in measurements. In total, 20 different cysteine pairs resulted in 52 unique samples that gave 54 total measured distances between the two Rad50 protomers in the MR^NBD complex (Supplementary file 1).

Following laser excitation of the Tb³⁺-chelate moiety and a 200 µs delay, donor-sensitized Bodipy FL or Cy3 fluorescence emission decay curves were collected for each of the MR^NBD LRET samples at 5°C in the presence of 5 mM Mg²⁺ and 2 mM ATP (Figure 1C). Under these conditions, Rad50^NBD should be >99% bound to ATP as the K_D for ATP is ~3 µM and there is no measurable ATP hydrolysis in 1 hr at 50°C. Collecting LRET data at 50°C (the maximum temperature for the fluorimeter) approaches the physiological temperature range for the hyperthermophilic Pf MR (>60°C) while minimizing Rad50-catalyzed ATP hydrolysis. In multiexponential fits, the emission decays were best described by two or three exponentials depending on the identity of the LRET pair (see Materials and methods). In all cases, the first lifetime (<100 µs) is a function of instrument response time and was discarded (Cooper and Altenberg, 2013; Zoghbi et al., 2017; Zoghbi et al., 2012). The Tb³⁺-chelate luminescence decays were also recorded at each probe position in donor-only-labeled MR^NBD complexes. As expected, the value of the Tb³⁺-chelate lifetime changed with the local environment of each cysteine. Using these Tb³⁺-chelate donor lifetimes, combined with the donor-sensitized acceptor lifetimes and the R₀ of the dye pair in the sample, distances were calculated between probes for each LRET pair.

For the majority of the 20 cysteine pairs analyzed, combining the data for the Bodipy/Tb³⁺- and Cy3/Tb³⁺-labeled samples gave three distinct distances. For 10 of the pairs (e.g., L51C-L51C), the longer distance (d2^Bodipy) in the Bodipy/Tb³⁺ samples matched the shorter distance (d1^Cy3) in the Cy3/Tb³⁺ samples, and the Cy3/Tb³⁺ samples gave a second, longer distance (d2^Cy3) (Figure 1D, Supplementary file 1). This longer distance became ‘visible’ in samples where Cy3 was the acceptor because the R₀ for Tb³⁺ and Cy3 is longer. For four of the pairs (e.g., L51C-A66C), the one Bodipy/Tb³⁺ distance did not overlap with the two Cy3/Tb³⁺-determined distances, and the combined data resulted in three distances. For two pairs (A66C-A66C and S93C-S93C), only one distance was seen in the Bodipy/Tb³⁺ data, while the Cy3/Tb³⁺ data contained two distances. In these samples, the d^Bodipy matched d1^Cy3 for a total of two distances. And finally, for four pairs (e.g., S13C-S93C) only Cy3/Tb³⁺ samples were made, resulting in two distances. Together, these data illuminate the presence of a third solution state in addition to the closed and partially open conformations.

To confirm that the distances observed in MR^NBD were the same in full-length MR, all of the cysteine mutations were also introduced into full-length Pf Rad50, which contains the long coiled-coil domains and apical zinc hook motif. We previously reported that the two native cysteines in the zinc hook motif of full-length Rad50 are not efficiently labeled by the LRET probes and do not result in LRET donor-sensitized acceptor signal (Boswell et al., 2020). Tb³⁺-chelate donor-only lifetimes measured in these mutants were identical to those measured for the Rad50^NBD construct, indicating that the local environments of the introduced cysteines do not change between full-length and NBD constructs. Unfortunately, because full-length Rad50 dimerizes at the zinc hook, cysteine mutants could not be mixed and only ‘identity’ LRET pairs (e.g., Bodipy-L51C and Tb³⁺-L51C) could be made. Nonetheless, for all LRET probe positions measured, the distances between full-length MR cysteine pairs were within a few Ångströms of those measured in MR^NBD (Figure 1D).

The measured LRET distances were input as unambiguous restraints in the HADDOCK molecular docking program (Dominguez et al., 2003; van Zundert et al., 2016), defining the Cβ-Cβ distance between the LRET-labeled residues. The unambiguous restraints included symmetrical distances for each LRET pair (e.g., both Rad50a protomer L51 to Rad50b protomer S13 and Rad50a S13 to Rad50b L51). Given the relatively large size and unknown orientation of the fluorophores and their linkers with respect to the protein, ±5 or ±7 Å bounds (for distances less than and greater than 75 Å, respectively) were used for the unambiguous LRET distance restraints in HADDOCK. These unambiguous restraints were used to dock two Rad50^NBD protomers with one Mre11 dimer in a three-body docking simulation.

The ‘closed’ HADDOCK model fit very closely to the measured LRET distances (Figure 2A, Figure 2—video 1, Supplementary file 2). HADDOCK returned 190 structures in five clusters, with 164 in the top-scored cluster. The Rad50 dimer formed in this model is nearly identical to the AMPPNP-bound dimer structure of Pf Rad50^NBD (PDB: 3QKU, all-atom root-mean-square deviation [RMSD] = 1.11 Å) (Williams et al., 2011). Except in two cases, the Cβ-Cβ distances between the Rad50 LRET pairs in the HADDOCK model were within ±5 Å of their respective unambiguous LRET restraint. LRET pairs A66-A66 and A66-S93 deviated more significantly with differences of 6.6 and 11.0 Å, respectively. This deviation could arise from slight differences in the loop structures between the solution (LRET) and crystal states (i.e., input PDB; note that HADDOCK does not move backbone atom positions during model refinement). Specifically, we observed the expected relative positions within the associated Rad50s for residues in the Walker A motif (N32) of one protomer and the D-loop (D829) and signature motif (S793) of the other (Figure 2D, top, Figure 2—video 1). In this closed model, Rad50 interacts with Mre11 via the capping domain and along the top of the nuclease domain as observed in the Thermotoga maritima (PDB: 3THO; Möckel et al., 2012), Methanocaldococcus jannaschii (PDB: 3AV0; Lim et al., 2011), and E. coli (PDB: 6S6V; Käshammer et al., 2019) nucleotide-bound structures (with Cα RMSDs of 4.9 Å, 1.9 Å, and 4.7 Å, respectively) (Figure 2—figure supplement 1A). Like the M. jannaschii structure, each Rad50 protomer makes contact with only one of the Mre11 capping domains mainly through interactions between capping domain β18 and Rad50 Lobe II αE and β8–10. In total, eight and four ionic or hydrogen bond interactions are made between Rad50 and the Mre11 capping and nuclease domains, respectively. In particular, unique contacts not seen in the other two conformations are made between Rad50 E831 and Mre11 H17 in the nuclease domain and Rad50 E758/E761 and Mre11 Y325/K327 in the capping domain. The combination of these interactions occludes the Mre11 nuclease active site for dsDNA, as previously described (Lim et al., 2011; Möckel et al., 2012).

The ‘partially open’ HADDOCK simulation returned nine clusters of models, and analysis of the top four clusters concluded that the distances between LRET pairs for each cluster had an average standard deviation of ~0.3 Å when compared among the top three clusters and increased to ~0.7 Å when adding the fourth (Figure 2B, Figure 2—video 1, Supplementary file 2). 13 out of 20 of the Cβ-Cβ distances in the HADDOCK model are within ±5.7 Å of the measured LRET distances (Supplementary file 2). Interestingly, all of the pairs with larger (>5.7 Å) deviations included either A66 or S93, again suggesting that the position of those loops differs between the crystal and solution conditions. In the partially open model, only 4.2 Å separates the Rad50 protomers at their closest point between the two D829 residues (Cα-Cα distance). In this conformer, the Walker A/D-loop interactions are no longer formed between the Rad50 protomers as the Walker A motifs have rotated out of the Rad50-Rad50 interface (Figure 2D, bottom, Figure 2—video 1). Instead, the D-loop and signature motifs face one another and are stacked in the interface. Although there are still significant interactions between the Rad50 protomers and the Mre11 capping and nuclease domains, several of which are maintained from the closed conformation, the partially open conformation has rotated by 22° to interact differently with the capping domain compared to the closed structure (Figure 2—figure supplement 1B). Specifically, interactions with the W308-D313 loop in the Mre11 capping domain are similar between the two conformations, but the Rad50 interactions with β18 have been broken in partially open, and Rad50 is now interacting with Mre11 β16 and β17 residues instead. Moreover, the movement of Rad50 has allowed the Mre11 capping domains to rotate inward toward the nuclease domains (Figure 2E). We observe a total of nine ionic or hydrogen bond interactions between the Mre11 capping domain and Rad50 and five between the nuclease domain and Rad50. In the capping domain, unique interactions occur between Mre11 K277 and Rad50 E750 and Mre11 R303 and Rad50 E754. Even with these contacts between Rad50 and Mre11, DNA could access the nuclease active site of Mre11.

Finally, HADDOCK returned 14 clusters of models for the ‘open’ complex, and analysis of the top four clusters showed that the distances between LRET pairs for each cluster had an average standard deviation of ~0.7 Å among the top three clusters and ~1.2 Å when adding the fourth. 10 out of 14 of the Cβ-Cβ distances in the top cluster are within ±5.3 Å of the input LRET unambiguous restraints, whereas the remaining 4 are within ±9.3 Å (Figure 2C, Figure 2—video 1, Supplementary file 2). The Rad50 protomers have moved apart considerably (~41 Å D829-D829 Cα-Cα distance). Moreover, the orientation of the Rad50 protomers with respect to Mre11 has changed significantly (Figure 2—figure supplement 1C) having rotated 13° and the base of the coiled-coil translated ~23 Å away from the partially open model. Because of this rotation and translation, there are minimal contacts (only two) with the Mre11 capping domain, which now occur between Rad50 β10 and Mre11 β16 and the C-terminus of this construct. Now, six ionic interactions are formed between the Mre11 nuclease domain and Rad50. Within the nuclease domain, unique interactions occur between Mre11 R177 and E181 on helix αE with Rad50 E841 and R842 and between Mre11 E152 on helix αD and Rad50 K860. In the capping domain, Mre11 K279 makes a unique interaction with Rad50 E783. With Rad50 rotated fully away, the capping domains move even closer to the nuclease domain and both Mre11 nuclease active sites are now fully accessible to dsDNA for exonuclease activity.

To ensure that the unambiguous distance restraints obtained from one probe position were not dominating the structure calculations, HADDOCK runs were performed with systematic dropouts of all restraints calculated from a specific cysteine position. For example, for the L51C probe position, L51C-S13C, L51C-L51C, L51C-A66C, L51C-S93C, L51C-N774C, and L51C-V866C distances were all removed from the unambiguous restraints, and HADDOCK runs were repeated for each of the three sets of distances (closed, partially open, and open). In general, none of the dropouts appreciably changed the overall conformation of any of the states (Figure 2—figure supplement 2).

Next, to decrease the stability of one or two conformations over the others, charge reversal mutations were made to several Mre11 residues that directly interact with Rad50. As there are a handful of shared interactions between Mre11 and Rad50 in the various conformation combinations, it was impossible to completely disrupt a given state. The Mre11 mutants were combined with full-length Rad50 to make MR complex, and then both Mre11 nuclease activity and Rad50 ATP hydrolysis activity were tested (Figure 3B–E). For Mre11 nuclease activity, three sets of fluorescence-based experiments were performed in 384-well plates (Figure 3B–D). The Exo2 assay interrogates the 3′-to-5′ Mn²⁺-dependent exonuclease activity as only 2 base-pairs need to be excised before the fluorescent 2-aminopurine nucleotide is released from the 3′-end of the double-stranded duplex (Figure 3B). The Exo11 signal results from a combination of exonuclease and endonuclease activity as the fluorescent nucleotide is 11 base-pairs from the 3′ end of the dsDNA and requires either sequential nuclease functions (e.g., endonuclease followed by exonuclease) or processive exonuclease activity to free it from the duplex (Figure 3C). Finally, the Cy3/BHQ2 ssDNA assay monitors endonuclease activity since the ssDNA needs to be cleaved to separate the quencher from the fluorophore (Figure 3D). In addition to these plate-based fluorescent assays, the cleavage of four different DNA substrates was monitored on 15% denaturing polyacrylamide gels: a 40-nucleotide ssDNA, a 40-nucleotide dsDNA, a 36-nucleotide hairpin DNA construct that includes a 2-nucleotide 3′-overhang, and a 50-nucleotide dsDNA with five phosphorothioate bonds that block 3′-to-5′ exonuclease activity at the 3′ end of the labeled strand (Figure 3—figure supplement 1). MR complexes made with Pf Mre11 H52S (exonuclease inactive) and H85S (exo- and endonuclease inactive) mutants were used as controls in all of these assays (Hopfner et al., 2001; Williams et al., 2008). In total, when the results of these various experiments were compared, we found that the Exo2 data mirrored the results of the 40-mer dsDNA cleavage in the gels in the absence of ATP – the 3′ nucleotide of the dsDNA was removed, but no further cleavage products were detected (Figure 3B, Figure 3—figure supplement 1A). The Exo11 results, on the other hand, very closely resembled those of the dsDNA 40-mer and hairpin DNA gels in the presence of ATP (Figure 3, Figure 3—figure supplement 1B C). Interestingly, the cleavage pattern of the blunt-ended dsDNA, but not the hairpin with the 3′ overhang, showed intense bands where the 3′ nucleotide had been cleaved off, and both of these substrates had a ‘ladder’ of cleavage products. Thus, it appears that the 3′-overhang of the hairpin dsDNA either precludes the 3′-to-5′ exonuclease activity of the complex or MR exonuclease activity occurs at the fifth nucleotide from the end. Unexpectedly, whereas the nuclease-inactive H85S showed no activity on any of the four gel-based DNA substrates, the previously described exonuclease-impaired H52S did show activity on all of these substrates. Note, to the best of our knowledge, that there are no reports in the literature of the nuclease activity of H52S on any substrate other than the Exo2 in the absence of ATP; therefore, the cleavage products observed with that mutant were surprising and perhaps further demonstrate the importance of endonuclease activity in Pf MR (Figure 3—figure supplement 1). Finally, the Cy3/BHQ2 ssDNA results were nearly identical to the results for the cleavage of the 40-nucleotide ssDNA in the gel as well as to the 3′-end phosphorothioate-blocked dsDNA (Figure 3D, Figure 3—figure supplement 1D E). Although both of these were worse substrates for MR as compared to the 40-mer dsDNA and the hairpin, they were both cleaved in the absence of ATP. Below, to describe the effects that each destabilizing Mre11 mutant has on MR nuclease activity, we used the results of the quantitative Exo2, Exo11, and Cy3/BHQ2 fluorescence plate assays (Figure 3).

In the closed conformation, Mre11 nuclease domain residue H17 interacts with Rad50 E831 (Figure 3—figure supplement 2A). MR H17E had ~30% of wild-type exonuclease activity on Exo2 DNA but does not cleave Exo11 DNA (Figure 3B and C). Williams et al., 2008 identified H17 as a ‘wedge residue’ that helps to unwind the dsDNA helix. H17E is, however, a competent ssDNA endonuclease, showing nearly twice the activity of WT MR on the Cy3/BHQ2 ssDNA substrate in the absence of ATP (Figure 3D). This result is consistent with previous reports that dsDNA and ssDNA substrates use different binding sites on Mre11 (Rahman et al., 2021; Rahman et al., 2020). When ATP is added, the ssDNA endonuclease activity of MR H17E decreases to WT levels. Although MR H17E cannot cleave dsDNA, it showed robust activity when the 3′-end of the DNA duplex is blocked with phosphorothioate bonds (Figure 3—figure supplement 1E). Moreover, the banding pattern of the 3′-phosphorothioate-blocked substrate was different for MR H17E when compared to wild-type or the other mutants. Therefore, we suggest that the cleavage products observed on this modified dsDNA for the H17E mutant were all from endonuclease activity, whereas the products of the other MR complexes originate from a combination of endo- and exonuclease activities. Finally, MR H17E decreased the V_max of ATP hydrolysis to ~60% of wild-type MR (Figure 3E), confirming that this residue assists in destabilizing the closed conformation from which hydrolysis proceeds.

Mre11 Y325 and K327 are located in the capping domain on β18 and interact with E761 and E758 of Rad50, respectively, in the closed conformation (Figure 3—figure supplement 2A). In the presence of ATP, the Y325A/K327E double mutant had ~85% of the exonuclease activity of wild-type MR, but only ~50% of the activity for the Exo2 substrate without ATP (Figure 3B and C). MR Y325A/K327E had an approximately twofold decrease in endonuclease activity on the Cy3/BHQ2 ssDNA substrate (Figure 3D). Additionally, MR Y325A/K327E decreased the V_max of ATP hydrolysis to the same level as MR H17E (Figure 3E). Thus, like MR H17E, destabilizing the closed conformation through the Mre11 Y325A/K327E mutant had the predictable effect of decreasing Rad50 ATP hydrolysis.

Mre11 D313 contacts Rad50 K785 in both the closed and partially open conformations (Figure 3—figure supplement 2A B). D313 and its neighboring residues in a loop at the top of the capping domain might be acting as a pivot point for Rad50 to rotate between the two states. Surprisingly, MR D313K had significantly increased nuclease activity on all of the substrates (Figure 3, Figure 3—figure supplement 1). In fact, D313K showed shorter cleavage products, indicating more nuclease cleavage, in the gel-based assays when compared to wild-type or the other mutants (Figure 3—figure supplement 1A–C). An approximately twofold increase in activity was observed for Exo2 in the absence of ATP (Figure 3B); thus, we hypothesize that destabilizing both the closed and partially open conformations increases the population of the open conformation that accommodates dsDNA substrate. Activity against the Exo11 substrate increased ~2.5-fold in this mutant (Figure 3C), and the Cy3/BHQ2 ssDNA results showed an ~1.3- and 3-fold increase in cleavage over WT in the absence and presence of ATP, respectively (Figure 3D). In contrast to its high levels of nuclease activity, MR D313K reduced the V_max for ATP hydrolysis by more than 50%, which was expected since the closed state is destabilized (Figure 3E). Therefore, Mre11 nuclease and Rad50 ATPase results for the D313K mutant demonstrate that the ATP hydrolysis function is not directly correlated with nuclease activity.

In the partially open complex, Mre11 K277 in β16 interacts with Rad50 E750, which is in αE at the base of a coiled-coil, whereas in the open complex, Mre11 K279 also in β16 interacts with Rad50 E783 in the short loop between β9 and β10 (Figure 3—figure supplement 2B C). As these two residues are close in sequence space, a double mutant was constructed to destabilize both partially open and open conformations. The most striking feature of MR K277E/K279E was that it had no nuclease activity on Exo2 without ATP but increased to ~50% of wild-type levels when ATP was added (Figure 3B). No other mutant tested required ATP for the Exo2 substrate, implying that it requires hydrolysis to open the complex before dsDNA can bind. Moreover, this mutant displayed the lowest activity in the presence of AMPPNP, suggesting that the combination of the mutations and non-hydrolyzable analog effectively stabilized the closed state, fully occluding the dsDNA substrate. On the Exo11 substrate, the activity was also ~50% of wild-type (Figure 3C). On the Cy3/BHQ2 ssDNA, the endonuclease activity decreased to ~20% of wild-type (Figure 3D). In fact, MR K277E/K279E had the worst endonuclease activity of all of the mutants tested, except for the nuclease-deficient H85S, suggesting that the closed state is not productive for endonuclease activity. This double mutant has no effect on ATP hydrolysis activity (Figure 3E) since the closed conformation can readily form.

Finally, mutants were made for Mre11 R177, E181, and E152, which all make unique contacts in the open conformation (Figure 3—figure supplement 2C). These residues are along the ‘top edge’ of the Mre11 nuclease domain in αD and αE, and R177/E181 and E152 are structurally homologous to part of the so-called ‘latching loop’ and ‘fastener,’ respectively, of the E. coli SbcCD cryo-EM ‘cutting state’ structure (Käshammer et al., 2019). The MR R177E/E181R double mutant had ~60% of wild-type nuclease activity (Figure 3B and C) and 60% (no ATP) to 80% (with ATP) Cy3/BHQ2 ssDNA endonuclease activity (Figure 3D), but wild-type ATP hydrolysis activity (Figure 3E), which is consistent with there being less open and more closed conformation. On the other hand, MR E152K, which interacts with Rad50 K860 in αE, showed more impaired nuclease activity than R177/E181, but nearly WT endonuclease activity, and, surprisingly, decreased the V_max for ATP hydrolysis by more than 50% (Figure 3). As this mutant should readily form the closed conformation, this result was unexpected and, like D313K, demonstrates the independence of nuclease activity on ATP hydrolysis. Similar to the K227E/K279E mutant, both MR R177E/E181R and MR E152K had very little Exo2 activity in the presence of AMPPNP, again suggesting a complete stabilization of the closed conformation.

The LRET distances used for the HADDOCK structural model calculations above were determined under saturating ATP conditions. Next, LRET measurements were collected for a subset of the LRET pairs without any nucleotide or in the presence of the non-hydrolyzable ATP analog ATPγS. LRET pairs S13-S13, L51-L51, and N774-N774 in both the MR^NBD and full-length MR constructs were examined. These three residues are distributed across the dimerization face of Rad50 (Figure 1B), and together should illuminate any significant changes in distance between the two Rad50 protomers of the MR complex. MR^NBD S13-L51 and N774-L51 LRET pairs were also examined. As reported in Table 1, three distances were observed for each LRET pair in the absence of nucleotide (‘apo’), and these distances did not differ substantially from the distances observed under saturating ATP conditions. Although the ‘open’ distances all increased, most were within the error of the distance observed with ATP and all were within the ±7 Å of the HADDOCK inputs. When the non-hydrolyzable ATP analog ATPγS was added, three distances were again observed for each LRET pair, these distances did not vary more than 3 Å from the distances recorded in the ATP samples, and nearly all the distances were within the error of the measurement made with ATP. Thus, any changes in MR structure due to binding or hydrolyzing ATP are limited to local conformations invisible to LRET as we have employed it. Additionally, the LRET data on the ATPγS-bound MR demonstrate that the partially open state is a stable conformation of the MR complex and does not appear to be an intermediate in the Rad50 ATP hydrolysis catalytic cycle.

To examine whether or not the addition of DNA substrates affected any of the MR conformations, either a hairpin DNA, with 15 base-pairs and a 2-nucleotide 3′-end overhang, or a 36-nucleotide ssDNA were added to the LRET samples containing 2 mM ATP. The only consistent difference in LRET distances was observed in the ssDNA samples: the open distances were shorter than what was noted in all other conditions (Table 1), indicating that the two Rad50s in the complex may be slightly closer together. None of the changes in LRET distances in the apo, ATPγS, or DNA-bound samples were large enough to necessitate recalculating HADDOCK. Note that our LRET measurements would not be able to detect if the Rad50 protomers were moving relative to the Mre11 dimer, as observed by cryo-EM in the E. coli SbcCD (MR homolog) ‘cutting’ state (Käshammer et al., 2019), or if the Mre11 protomers were moving relative to each other (Käshammer et al., 2019; Saathoff et al., 2018). Unfortunately, it is likely impossible to measure LRET between Mre11 and Rad50 within an MR complex because of cross-talk between the different protomers (i.e., Rad50a to Mre11a, Rad50b to Mre11b, Rad50a to Mre11b, and Rad50b to Mre11a) within the three states.

The different nucleotide-bound states and DNA substrates are most likely altering the population distribution of the various conformations of the MR complex. Unfortunately, it is not feasible to assign relative populations to these structures as we may not have a complete inventory of states, we do not know to what extent other structures (e.g., extended or ‘cutting’ state) are populated, and each LRET decay curve only provides population information for the one or two conformations that are described by the data fitting.

To further validate the multiple conformations indicated by the LRET data, SAXS data was collected for samples of MR^NBD with and without ATP. The SAXS profiles for wild-type MR^NBD were compared to our three models using the FoXS webserver (Schneidman-Duhovny et al., 2016; Schneidman-Duhovny et al., 2013), which calculates the SAXS profile from an input structural model and compares that to an experimental SAXS profile. The ATP-bound MR^NBD experimental SAXS profile fit best to the partially open model (χ² = 0.34) followed by the closed model (χ² = 0.51), based on the single-state χ² values (Figure 4). The open model did not fit well to the experimental SAXS data (χ² = 13.6). Conversely, for the ATP-free MR^NBD SAXS profile, a reasonable fit was only obtained with the open model (χ² = 1.26), although a discrepancy between the experimental and back-calculated scattering data was observed for scattering angles of ~0.1–0.2 Å^–1 (Figure 4A, left). Next, MultiFoXS was used to fit the apo and ATP-bound MR^NBD SAXS profiles to either two or three of the LRET-derived models simultaneously (Figure 4A, right, and B). The fit of the ATP-bound SAXS profile to two models improved the χ² significantly and combined populations of open (19%) and closed (81%) conformations (Figure 4B). The fit using all three models did not improve the χ² but did assign populations of 15% open, 18% partially open, and 67% closed. The two-state fit to the apo MR^NBD SAXS profile only slightly improved the χ² (1.26 vs. 1.05) and gave 90% open and 10% closed, while the three-state fit gave 89% open, 5% partially open, and 6% closed again without improving χ² further. Here, minor improvements in the fit were observed in the 0.1–0.2 Å^–1 range; however, the quality was still not as good as the multistate fits to the ATP-bound SAXS data, implying that an additional, unobserved state might be populated in apo MR^NBD. Thus, the SAXS data generally support the presence of multiple states as observed in LRET data and demonstrate the expected shift in population to the closed form in the presence of ATP.

Plasmids for P. furiosus full-length Rad50, Rad50 NBD (Rad50^NBD: aa1-195; GGAGGAGG linker; aa709-882), and Mre11 protein expression in E. coli were previously described (Boswell et al., 2020; Boswell et al., 2018). Point mutations were introduced using a modified QuikChange (Strategene) approach and were verified by Sanger sequencing. Protein expression and purification were performed as previously described for full-length Rad50 (Boswell et al., 2020), Rad50^NBD (Boswell et al., 2018), and Mre11 (Rahman et al., 2020).

Rad50 ATP hydrolysis assays were performed essentially as described by Boswell et al., 2018. 0–300 µM ATP was titrated into microfuge tubes containing either 2.5 µM MR^NBD complex or 2 µM full-length MR complex and 50 mM Tris, 80 mM NaCl, 1% glycerol, 5 mM MgCl₂, pH 7. Reactions without protein were included for each ATP concentration to control for ATP degradation and PO₄ contamination. 60 µL reactions were incubated at 65°C for 60 min after which the tubes were placed on ice. 50 µL of each reaction was then transferred to the wells of clear, flat-bottom 96-well plates, and 100 µL of cold BIOMOL Green (Enzo Lifesciences) colorimetric reagent was added. After a 30 min incubation at room temperature to allow the color to develop, the amount of inorganic phosphate released by hydrolysis was quantified using the absorbance mode on a Synergy Neo2 multi-mode plate reader. BIOMOL Green signal (A₆₄₀) was corrected by subtracting the A₆₄₀ values of the ATP-only reactions at each ATP concentration and then transformed into pmols of PO₄ released/min based on a PO₄ standard curve incubated in BIOMOL Green reagent for 30 min at room temperature. Plots of PO₄ released/min (v₀) versus ATP concentration were fit to the Michaelis–Menten equation including a Hill coefficient (n).(1)v0=VmaxATPnKMn+ATPn

Nuclease activity was assayed by monitoring the fluorescence of a 2-aminopurine (2-AP) nucleotide incorporated into a dsDNA substrate. 2-AP fluorescence is quenched within the base-pair, and release of the 2-AP from the duplex by Mre11 nuclease activity results in an increase in the 2-AP fluorescence signal. The Exo2 substrate was formed by annealing equimolar amounts of 5′-GGCGTGCCTTGGGCGCGCTGCGGGCGG[2-AP]G-3′ and 5′-CTCCGCCCGCAGCGCGCCCAAGGCACGCC-3′ DNA oligos (IDT). For LRET Rad50^NBD cysteine mutants, 30 µL exonuclease reactions were mixed in microfuge tubes and contained 0.5 µM MR^NBD complex and 1 µM Exo2 dsDNA substrate in 50 mM HEPES, 100 mM NaCl, 5 mM MgCl₂, 0.1 mM EDTA, 1% glycerol, 1 mM TCEP, pH 7. 1 mM MnCl₂ was added to the reactions as indicated. After a 45 min incubation at 60°C, the tubes were removed from the heat block and spun. 25 µL of each reaction was transferred to black, flat-bottom 384-well plates and 2-AP fluorescence was quantified by the Synergy Neo2 plate reader (ex310/em375). For the Mre11-Rad50 interaction mutants, exonuclease activity was assessed for MR complexes made with full-length Rad50. Reactions with 0.5 µM full-length MR complex and 1 µM Exo2 were assembled as above, except 1 mM MnCl₂, 1 mM MnCl₂/1 mM ATP, or 1 mM MnCl₂/1 mM AMPPNP were added as indicated and reactions were incubated at 60°C for 15 min. In addition to the Exo2 substrate, a second substrate, Exo11, was also employed for full-length MR: 5′-GGCGTGCCTTGGGCGCGC[2-AP]GCGGGCGGAG-3′ annealed to 5′-CTCCGCCCGCTGCGCGCCCAAGGCACGCC-3′ (IDT). Exo11 reactions were incubated at 60°C for 30 min. Endonuclease activity assays of the various Mre11-Rad50 interaction mutants used a 17-nucleotide ssDNA substrate containing a fluorophore and quencher pair (5′-Cy3-TCTCTAGCAGTGGCGCC-BHQ2-3′; IDT), based on a previously reported assay (Yuan et al., 2016). Once the ssDNA is cleaved, the fluorophore and quencher separate, allowing Cy3 fluorescence to be detected. Reactions with full-length MR complexes were performed just as the exonuclease assays described above, except 0.2 µM Cy3/BHQ2 substrate was used. Endonuclease reactions were incubated for 30 min at 60°C. Cy3 fluorescence was then quantified by the Synergy Neo2 plate reader (ex535/em570). Data for each of the described nuclease assays were collected in triplicate.

Nuclease activity of full-length MR complexes on four separate substrates was analyzed on denaturing polyacrylamide gels: a 40-nucleotide ssDNA substrate (5′-Cy3-GTGTTCGGACTCTGCCTCAAGACGGTAGTCAACGTGCTTG-3′; IDT), a 40-nucleotide dsDNA substrate (the ssDNA substrate annealed to an unlabeled complementary strand), a 36-nucleotide DNA hairpin (5′-FAM-CACGCACGTAGAAGCTTTTGCTTCTACGTGCGTGAC-3; IDT, containing a 15-base-pair helix and a 2-nucleotide 3′-overhang), and a 50-nucleotide dsDNA where the labeled strand has phosphorothioate bonds between the six nucleotides at the 3′-end (5′-Cy5-CTGCAGGGTTTTTGTTCCAGTCTGTAGCACTGTGTAAGACAGGCCsAsGsAsTsG-3′, annealed to 5′- CATCTGGCCTGTCTTACACAGTGCTACAGACTGGAACAAAAACCCTGCAG-3′; IDT). Nuclease reactions in microfuge tubes contained 0.5 µM full-length MR complex and 1 µM DNA substrate in 50 mM HEPES, 100 mM NaCl, 5 mM MgCl₂, 1 mM MnCl₂, 0.1 mM EDTA, 1% glycerol, 1 mM TCEP, pH 7. 1 mM ATP was added to the reactions as indicated. After a 45 min incubation at 60°C, 1.5 µL of each reaction was transferred to 28.5 µL of the above buffer with 20 mM EDTA. These samples were then mixed with 30 µL of loading buffer (8 M urea, 20 mM EDTA, 6% Ficoll 400) before 20 µL was loaded onto 15% denaturing polyacrylamide gels in 1× TBE buffer. Gels were run at 3 hr at a constant power of 20 W and scanned by an Amersham Typhoon 5 fluorescence imager. FAM-labeled substrates were imaged with a 488 nm laser and 525BP20 filter, Cy3-labeled substrates with a 532 nm laser and 570BP20 filter, and Cy5-labeled substrates with a 635 nm laser and 670BP30 filter. The images were analyzed using ImageQuant software.

Thiol-reactive Tb³⁺ chelate DTPA-cs124-EMPH (LanthaScreen, Life Technologies, Inc) was used as the LRET donor for all samples and Bodipy FL maleimide (Invitrogen) or Cyanine3 (Cy3) maleimide (GE Healthcare) were used as acceptor fluorophores. 50 µM purified Rad50^NBD containing a single cysteine was combined with a twofold excess of one of the labels in degassed Labeling Buffer (25 mM Tris, pH 8, 100 mM NaCl, 10% glycerol, 1 mM TCEP) in 100 µL reactions. The reaction was incubated in the dark at room temperature for 1.5–2 hr. Labeling of full-length Rad50 cysteine mutants followed the same protocol, except the donor and one acceptor label were added simultaneously to each reaction (twofold molar ratio of each). Unreacted label was removed by running the labeling reaction over a Superdex 200 Increase 10/300 GL column (GE Healthcare) equilibrated with 25 mM HEPES, 200 mM NaCl, 10% glycerol, 1 mM TCEP, pH 7. Eluted protein was concentrated using centrifugal concentrators (VivaSpin, Sartorius) and then assessed for successful labeling by measuring fluorescence using the following settings and the monochromator in the BioTek Synergy Neo2 plate reader: ex337/em490 for DPTA-Tb³⁺, ex485/em515 for Bodipy FL, and ex535/em570 for Cy3. Concentrated labeled proteins were aliquoted, frozen in liquid nitrogen, and stored at –80°C.

For LRET experiments, 40 µL mixtures of 1 µM Mre11 + 0.5 µM Tb³⁺-labeled Rad50^NBD + 0.5 µM Bodipy- or Cy3-labeled Rad50^NBD were heated at 60°C for 15 min to form the MR^NBD complex. Full-length MR LRET mixtures were 1 µM Mre11 + 1 µM dual-labeled (donor + acceptor) Rad50. Each LRET sample was subsequently diluted to 160 µL (0.25 µM complex) with LRET buffer (50 mM HEPES, 100 mM NaCl, 5 mM MgCl₂, 0.1 mM EDTA, 1% glycerol, 1 mM TCEP, pH 7). Next, for samples containing nucleotide, 2 mM of either ATP or ATPγS was added and the samples were heated again for 20 min at 50°C. Finally, for samples containing DNA, either 34 µM hairpin DNA (5'-CACGCACGTAGAAGCTTTTGCTTCTACGTGCGTGAC-3′) or 16.6 µM ssDNA (5'-TGTAGTGCATTGCGTTTTTGCTTCTACGTGCGTGAC-3') was added to ATP-bound LRET samples and heated once more for 15 min at 50°C. These concentrations of DNA should give >95% bound complex in LRET conditions at 50°C. 150 µL were transferred into a 3 mm pathlength cuvette, and LRET data were collected after a final 5 min incubation in the fluorimeter at 50°C. Donor and acceptor lifetimes were calculated from intensity decays measured with an Optical Building Blocks phosphorescence lifetime photometer (EasyLife L). LRET samples were excited through a narrow band 335 nm filter (Semrock FF01-335/7) and after a 200 µs delay (to allow the decay of sample autofluorescence, emission due to direct excitation of the acceptor, and scattering of the excitation pulse), donor emission intensity was collected for Tb³⁺ at 50 Hz through a 490/10 nm band-pass filter (Omega Optical), and donor-sensitized acceptor emission intensities were collected at 100 Hz through 520/10 nm (Bodipy FL) or 570/10 nm (Cy3) band-pass filters. PTI FeliX32 software was used to fit the Bodipy FL and Cy3 emission decay curves to either a two- or three-exponential function depending on the identity of the LRET pair. For samples where the distance between the LRET pair in the nucleotide-bound crystal structure was ≤42 Å, the data fit well to three exponentials for both Bodipy- and Cy3-labeled samples. For samples where the LRET pair was separated by more than 45 Å, the Bodipy emission decays fit to only two exponentials while the Cy3 decays fit to three. Lifetime distributions shorter than ~100 µs were discarded as these are largely a function of the instrument response time (Cooper and Altenberg, 2013; Zoghbi et al., 2017; Zoghbi et al., 2012; Zoghbi and Altenberg, 2018). Donor Tb³⁺-chelate fluorescence decays were recorded at each probe position in donor-only-labeled MR^NBD complexes. Each Tb³⁺-chelate fluorescence decay fit well to two exponentials, with the longer lifetime comprising >85% of the signal. This longer lifetime was used in the distance analysis. From the Tb³⁺-chelate, Bodipy FL, and Cy3 lifetimes, the distances between donor and acceptor molecules (R) were calculated with(2)E=1-τDAτD

and(3)R=R0(E−1−1)1/6

where E is the efficiency of energy transfer, τ_DA is the donor-sensitized lifetime of the acceptor (Bodipy FL or Cy3), τ_D is the lifetime of the donor (Tb³⁺), and R₀ is the Förster distance between Tb³⁺ and Bodipy FL (44.9 Å) or Tb³⁺ and Cy3 (61.2 Å). Errors are the standard deviations from the mean of at least three measurements.

Molecular docking of the MR^NBD complex was done using the GURU interface on the HADDOCK 2.4 webserver (Dominguez et al., 2003; van Zundert et al., 2016). For the three-body docking protocol, the PDB inputs were 3DSC (Pf Mre11 dimer lacking the helix-loop-helix [HLH] motifs and C-termini, DNAs deleted) (Williams et al., 2008) and two monomers of 3QKU (AMPPNP-bound Pf Rad50^NBD in complex with the Mre11 HLH motif) (Williams et al., 2011). Except for increasing the number of structures for rigid body docking (it0) to 3000, all settings used were default. The three linkers attaching the Pf Mre11 capping domain to the nuclease domain were allowed to be fully flexible (Y222-V236, Y249-G254, and V266-F273). To allow HADDOCK to move the two Rad50 monomers relative to each other, we input Rad50 as two identical monomers. C2 symmetry was enforced between the two Rad50^NBD monomers and between the two monomers of the Mre11 dimer.

HADDOCK Mre11 to Rad50 active ambiguous interaction restraints (AIRs) were based on M. jannaschii (3AV0) (Lim et al., 2011) and E. coli (6S6V) (Käshammer et al., 2019) ATP-γ-S-bound MR structures. Passive AIRs (solvent accessible surface neighbors of active residues) were automatically defined by HADDOCK based on these active AIRs. Two extra passive AIRs were included on the ‘back side’ of the Mre11 capping domain to allow for the extended structure seen in ATP-free T. maritima MR (PDB: 3QG5; Lammens et al., 2011). Figure 1—figure supplement 2 shows the location of these AIRs on the structures of Mre11 and Rad50. The same set of active and passive AIRs were used in all of the HADDOCK runs with 50% random exclusion of AIRs in each structure calculation. The measured LRET distances were input as unambiguous restraints and defined as the Cβ-Cβ distance between the LRET-labeled residues, ± 5 Å. For distance restraints greater than 75 Å, the bounds were increased to ±7 Å as the lifetime fits giving these distances were in a more error-prone region (i.e., the flat section) of the Tb³⁺-Cy3 LRET efficiency curve. The model with the lowest HADDOCK score in each run was considered as the best structure. To ensure that the unambiguous distance restraints obtained from one probe position were not dominating the structure calculations, HADDOCK runs were performed with systematic dropouts of all restraints calculated from a specific cysteine position (Figure 2—figure supplement 2). For the closed model, all of the dropout structures maintained the proper D-loop/Walker A juxtaposition. PyMOL version 2.4 was used to calculate RMSD values for all-atom alignments of HADDOCK models and to make figures of structures.

MR^NBD samples for SAXS were made by mixing Mre11 and Rad50^NBD in a 1:1.1 molar ratio. Samples were heated at 60°C for 15 min to form complex and then cooled on the benchtop. The complex was then loaded onto a HiLoad 16/60 Superdex 200 column (GE Healthcare) equilibrated in 25 mM HEPES, 0.2 M NaCl, 0.1 mM EDTA, 1 mM TCEP, pH 7. The MR^NBD peak was collected and concentrated (Vivaspin, Sartorius) before dialysis (Slide-A-Lyzer MINI, Thermo Scientific) into LRET buffer (50 mM HEPES, 100 mM NaCl, 5 mM MgCl₂, 0.1 mM EDTA, 1% glycerol, 1 mM TCEP, pH 7) ± 2 mM ATP overnight at room temperature. Samples were diluted to 4 mg/mL (~22 µM MR^NBD complex) with the equilibrated dialysis buffer and sent to the SIBYLS beamline 12.3.1 at the Advanced Light Source in Berkeley for high-throughput SAXS analysis (Classen et al., 2013; Dyer et al., 2014; Hura et al., 2009). For each sample, a total of 33 frames were collected at 0.3 s intervals at 10°C with a cell thickness of 1.5 mm. The sample to detector distance was 2 m, and the beam wavelength was 11 keV/1.27 Å. Buffer profiles were subtracted from sample profiles, and the buffer-subtracted frames for each sample were averaged using SAXS FrameSlice (https://sibyls.als.lbl.gov/ran).

Compared to the 3QKU PDB used as the input for HADDOCK runs, the Rad50^NBD protein construct used in the LRET and SAXS samples has an additional seven amino acids on one side of the truncated coiled-coils and eight amino acids on the other and the linker (GGAGGAGG) connecting them is slightly longer. To ensure that the structural models were as close to the protein construct used for the SAXS experiments as possible, the Rad50 coiled-coil and linker along with the 14 amino acid loop connecting the Mre11 helix-loop-helix and nuclease domain were modeled with loop modeling via Rosetta (version 3.11) (Mandell et al., 2009; Wang et al., 2007). This resulted in models only missing the 47 C-terminal amino acids of Mre11 as compared to the experimental protein (Figure 4—figure supplement 1). The FoXS server was used to back-calculate SAXS profiles of these modified closed, partially open, and open HADDOCK-Rosetta models and to fit these to the experimental MR^NBD scattering curves. The MultiFoXS server was used to calculate population-weighted ensemble fits (Schneidman-Duhovny et al., 2016; Schneidman-Duhovny et al., 2013).