Heterogeneity in E. coli RecBCD Helicase-DNA binding and base pair melting

Abstract

E. coli RecBCD, a helicase/nuclease involved in double stranded (ds) DNA break repair, binds to a dsDNA end and melts out several DNA base pairs (bp) using only its binding free energy. We examined RecBCD-DNA initiation complexes using thermodynamic and structural approaches. Measurements of enthalpy changes for RecBCD binding to DNA ends possessing pre-melted ssDNA tails of increasing length suggest that RecBCD interacts with ssDNA as long as 17–18 nucleotides and can melt at least 10–11 bp upon binding a blunt DNA end. Cryo-EM structures of RecBCD alone and in complex with a blunt-ended dsDNA show significant conformational heterogeneities associated with the RecB nuclease domain (RecB^Nuc) and the RecD subunit. In the absence of DNA, 56% of RecBCD molecules show no density for the RecB nuclease domain, RecB^Nuc, and all RecBCD molecules show only partial density for RecD. DNA binding reduces these conformational heterogeneities, with 63% of the molecules showing density for both RecD and RecB^Nuc. This suggests that the RecB^Nuc domain is dynamic and influenced by DNA binding. The major RecBCD-DNA structural class in which RecB^Nuc is docked onto RecC shows melting of at least 11 bp from a blunt DNA end, much larger than previously observed. A second structural class in which RecB^Nuc is not docked shows only four bp melted suggesting that RecBCD complexes transition between states with different extents of DNA melting and that the extent of melting regulates initiation of helicase activity.

Introduction

E. coli RecBCD is a complex enzyme possessing both DNA helicase and nuclease activities [1, 2],[3] with two main functions. The first is to degrade foreign DNA that invades the host bacterium. The second is to initiate repair of double stranded DNA breaks in the host chromosome, which are lethal if unrepaired. The RecBCD enzyme possesses three subunits; RecB (134kDa) is a superfamily 1A helicase/translocase motor that moves along single stranded (ss) DNA with 3’ to 5’ directionality[4, 5] (Figure 1A). RecD is a superfamily 1B helicase/translocase motor that moves along ssDNA with 5’ to 3’ directionality[6–8]. During duplex DNA unwinding by RecBCD, the RecD and RecB motors translocate along the complementary ssDNA strands and thus move in the same net direction. RecB also possesses a 30 kDa nuclease domain at its C-terminus that is connected to the motor domain via a ~70 amino acid linker[9, 10]. The RecC subunit is structurally similar to RecB, but has no ATPase activity[11]. RecBCD also possesses a secondary translocase activity residing in RecBC that may reflect a double stranded DNA translocase [4, 7, 12, 13]. RecC contains the region that recognizes the chi (crossover hotspot instigator) sequence (5’-GCTGGTGG) which identifies the DNA as being from the host chromosome[14–16]. After chi recognition, major changes occur in the enzyme[17, 18]. Among them, the nuclease activity, which had been degrading both ssDNA strands during DNA unwinding, begins to degrade only the 5’-ended ssDNA, leaving the 3’-ended ssDNA available for loading of a RecA protein filament[19]. This RecA filament then begins a search for DNA homology to initiate a recombination event that results in repair of the double stranded DNA break.

Figure 1.

(A) Cutaway view of surface representation of a RecBCD-DNA crystal structure (1w36) [21]. RecB motor domain (red), RecB nuclease domain (pink), RecC (blue), RecD (green), DNA (yellow). Surface representation for DNA was made semi-transparent and overlaid with a ribbon cartoon representation of the DNA. (B) Continuous sedimentation, c(s), distribution of our RecBCD sample (200 nM) obtained from a sedimentation velocity experiment in Buffer M50–10, 25°C, showing a single symmetric peak at s_20,w=11.9. S (s=10.8 S before correction). (C) 8% SDS polyacrylamide gel of RecBCD stained with coomassie brilliant (right lane) and Benchmark™ protein ladder (left lane).

One of the more interesting, but poorly understood activities of RecBCD is its ability to recognize a double stranded DNA end and in an ATP-independent reaction, using only its binding free energy, form an initiation complex in which a number of base pairs (bp) are separated or “melted” at the duplex end [20–22]. Crystal structures[21, 23] and cryo-EM structures[24] have provided some details of the protein-DNA interactions in the initiation complex, indicating that at least 4–6 bp are melted (Figure 1A). However, the thermodynamic forces that drive complex formation have not been investigated. Previous studies have examined the effect of ssDNA flanking regions on the equilibrium binding of RecBCD to a duplex DNA end [25, 26]. Wong et al.[26]examined the affinity of RecBCD to DNA ends in the absence of Mg²⁺ using fluorescence approaches and demonstrated that the affinity, K_BCD, depends on the length of the ssDNA flanking regions (tails). For DNA ends possessing only a single 3’-(dT)_n tail, K_BCD increases up to n=6, but then decreases as n increases further. For DNA ends possessing only a 5’-(dT)_n tail, K_BCD increases and reaches a plateau at n=10. Hence, it was concluded that RecBCD binds with highest affinity to a DNA end possessing a pre-melted end with 3’-(dN)₆ and 5’-(dN)₁₀ ssDNA flanking regions. Kinetic studies of DNA unwinding from various DNA ends also showed that the ssDNA flanking regions affect the rate of initiation of DNA unwinding[27–31]. To better understand the energetics of RecBCD binding and DNA melting we examined the thermodynamics of RecBCD binding to a variety of DNA ends using isothermal titration calorimetry (ITC), allowing us to dissect the free energy of binding into its enthalpic and entropic contributions. To complement these studies, we also determined cryo-EM structures of RecBCD alone and in complex with a blunt DNA end. Both approaches show that RecBCD is capable of melting at least 11 DNA bp from a blunt DNA end, more than previously observed, but the structures show significant structural heterogeneity.

Results

Assembly State of RecBCD

Buffers are denoted as Buffer MX-Y where X and Y are the [NaCl] and [MgCl₂] in mM units (e.g., buffer M50–10 has 50 mM NaCl and 10 mM MgCl₂). Our RecBCD purification yields homogeneous samples that are > 98% RecBCD hetero-trimer containing all three full length subunits with no contamination from a hexameric form[13, 32]. Figure 1B shows a continuous sedimentation coefficient distribution (c(s)) profile from a sedimentation velocity experiment performed on our RecBCD sample in Buffer M50–10 at 25°C showing a single symmetric peak with s_20,w=11.9 S, indicating that the sample exists as a monodisperse hetero-trimer, also verified by sedimentation equilibrium experiments[26]. The sample does not contain either RecBC (s_20,w=11.2 S) nor RecBCD hexamer (s_20,w=20.3 S). An SDS polyacrylamide gel (Figure 1C) shows that all three full length subunits are present. We routinely verified the heterotrimeric nature of our RecBCD samples by performing sedimentation velocity experiments before each set of DNA binding experiments.

Dissecting the energetics of RecBCD-DNA binding from DNA base pair melting

Our approach to dissect the energetics of RecBCD binding from DNA base pair melting is summarized in Figure 2A. RecBCD binding to a blunt DNA must overcome the unfavorable energetic contributions associated with base pair melting. We can express the total free energy of RecBCD binding to a blunt end, ∆G°_blunt, as consisting of contributions from ∆G°_PD and ∆G°_melt as in Eq. (1a), where ∆G°_PD represents the favorable contributions due to the RecBCD-DNA interactions in the final bound state and ∆G°_melt represents the unfavorable contributions from base pair melting.

Figure 2. Thermodynamics of RecBCD-DNA binding.

(A) Cartoon depiction of RecBCD (green) binding to a blunt DNA end vs a fully pre-melted DNA end. The ∆H for binding to a blunt DNA end, ∆H_blunt = ∆H_PD + ∆H_melt, where ∆H_PD is the favorable enthalpy change due to the protein-DNA interactions and ∆H_melt is the unfavorable enthalpy change for base pair melting. The ∆H for binding to a pre-melted DNA, ∆H_PM = H_PD; the difference, ∆∆H = (∆H_blunt − ∆H_PM) = ∆H_melt. (B) and (C) Binding isotherm from ITC experiment in which RecBCD (520 nM) or RecB^D1080ACD (600 nM) in the cell was titrated with a 60 bp blunt DNA (5 µM) in buffer M50–10, 25°C, yielding ΔH_blunt=15.5±0.1 kcal/mol per DNA end for RecBCD and ΔH_blunt=15.7±0.1 kcal/mol DNA end for RecB^D1080ACD. Note that the total ∆H shown in panels (B) and (C) represents the ∆H for binding to two ends. In both cases, the binding affinity was too high to be measured accurately, with a lower limit of K >10⁹ M⁻¹. (D) Binding isotherm from ITC experiment in which RecB^D1080ACD (430 nM) in the cell was titrated with blunt DNA (3.3 or 4 µM) in Buffer M275–10, 25°C, yielding ΔH_blunt= 4.9±0.1 kcal/mol and K=1.2(±0.1)×10⁷M⁻¹. (E) Binding isotherm from ITC experiment in which RecB^D1080ACD (520 nM) in the cell was titrated with dT₂₅/dT₂₅-tailed DNA (4 µM) in Buffer M275–10, 25°C, yielding ΔH_dT25= −76±3 kcal/mol for RecB^D1080ACD binding to one end of dT₂₅dT₂₅ DNA. The binding affinity was too high to be measured accurately, with a lower limit of K >10⁹ M⁻¹. (F) ΔH_obs from ITC experiments for RecBCD (blue circles) and RecB^D1080ACD (filled red circles) binding to a series of DNA ends possessing twin dT_n/dT_n tails vs (n) in Buffer M275–10, 25°C. Results for RecB^D1080ACD binding to DNA ends in the buffer used for the cryo-EM experiments (open red circles). See Supplementary Figure 1 and Supplementary Table 2 for details. The dashed lines show that the initial linear dependence of ∆Hobs on n intersects the plateau value at n= 17.5. The ∆∆H = ∆H_blunt − ∆H_PM (plateau) = ∆H_melt=81±2kcal/mol.

$$\mathrm{\Delta }{{\text{G}}^{\text{o}}}_{\text{blunt}}=\mathrm{\Delta }{{\text{G}}^{\text{o}}}_{\text{PD}}+\mathrm{\Delta }{{\text{G}}^{\text{o}}}_{\text{melt}}$$ (1a)

Similarly, we can express the total enthalpy change as in Eq. (1b).

$$\mathrm{\Delta }{\text{H}}_{\text{blunt}}=\mathrm{\Delta }{\text{H}}_{\text{PD}}+\mathrm{\Delta }{\text{H}}_{\text{melt}}$$ (1b)

However, this unfavorable contribution due to base pair melting will be eliminated for RecBCD binding to a DNA end that is already fully pre-melted as indicated in Eqs. (2), where the subscript “PM” refers to RecBCD binding to a pre-melted DNA end.

$$\mathrm{\Delta }{{\text{G}}^{\text{o}}}_{\text{PM}}=\mathrm{\Delta }{{\text{G}}^{\text{o}}}_{\text{PD}}$$ (2a)

$$\mathrm{\Delta }{\text{H}}_{\text{PM}}=\mathrm{\Delta }{\text{H}}_{\text{PD}}$$ (2b)

With the assumption that the final state for RecBCD binding to a pre-melted DNA end is the same as the final state for RecBCD binding to a blunt end, we can obtain an estimate of the energetic contributions from bp melting by comparing the energetics of RecBCD binding to a blunt end vs. a fully melted end as in Eqs. (3).

$$\mathrm{\Delta }\mathrm{\Delta }{\text{G}}^{\circ }=\mathrm{\Delta }{{\text{G}}^{\circ }}_{\text{blunt}}-\mathrm{\Delta }{{\text{G}}^{\circ }}_{\text{PM}}=\mathrm{\Delta }{{\text{G}}^{\circ }}_{\text{melt}}$$ (3a)

$$\mathrm{\Delta }\mathrm{\Delta }\text{H}=\mathrm{\Delta }{\text{H}}_{\text{blunt}}-\mathrm{\Delta }{\text{H}}_{\text{PM}}=\mathrm{\Delta }{\text{H}}_{\text{melt}}$$ (3b)

Although we can measure both the binding affinity, and thus ∆G° as well as ∆H for RecBCD binding to a blunt DNA end by isothermal titration calorimetry (ITC), we can only measure ∆H for RecBCD binding to a fully pre-melted DNA end using ITC since the binding affinity is too high to measure accurately, hence we are only able to estimate ∆H_melt in our studies. However, since previous thermodynamic studies of DNA melting provide the range of values for ∆H for bp melting[33], we can use our estimate of ∆H_melt to estimate the number of bp that can be melted upon RecBCD binding to a blunt DNA end. This approach was used previously to examine RecBC melting of DNA ends with the conclusion that RecBC is capable of melting 6 bp upon binding to a blunt DNA end[22].

Thermodynamics of RecBCD binding to a blunt DNA end

We examined RecBCD binding to a 60 bp blunt-ended dsDNA using ITC, by titrating DNA into RecBCD. The sequence of the 60 bp DNA duplex is given in Supplementary Table 1. We have shown previously that RecBCD can bind independently to each end of this 60 bp DNA with identical affinities [26, 34]. Figure 2B shows the results of an ITC experiment in Buffer M50–10, 25°C. Under these conditions, the binding affinity is too high (binding is stoichiometric) so that we can only estimate a lower limit of K_obs >10⁹ M⁻¹. The sharp breakpoint at [DNA_tot]/[RecBCD_tot]= 0.5 indicates that two RecBCD trimers bind to each DNA duplex at saturation, one at each end, indicating that RecBCD is fully active in DNA binding. Under these conditions, we measure a binding enthalpy of ∆H= 15.5±0.1 kcal/mol. Note that the total ∆H = +31.0 ±0.2 kcal/mol shown in Figure 2B is for the binding of two RecBCD molecules. Since we cannot obtain information of K_obs (∆G⁰), we also cannot obtain the T∆S⁰ component of binding. However, because RecBCD binding to a blunt DNA end is enthalpically unfavorable, the binding must be driven by favorable entropic changes, even though we are unable to quantitate at this low [NaCl]. We also examined RecBCD binding to the 60 bp blunt DNA in the buffer used for our cryo-EM studies (20 mM Tris, pH 7.4, 50 mM NaCl, 4 mM MgCl₂) at 25.0°C. Under these conditions, RecBCD binding is also stoichiometric with ∆H_obs = 13.7±0.2kcal/mol, very similar to that measured in Buffer M50–10 (Supplementary Figure 1Ci). The large unfavorable ∆H observed for RecBCD binding to a blunt end is likely due in large part to the fact that RecBCD disrupts or “melts” some number of base pairs to form an initiation complex as demonstrated previously [20–22, 26].

In order to lower the equilibrium binding constant into a range that we can measure by ITC, we raised the [NaCl] to 275 mM. In Buffer M275–10, 25.0°C (Supplementary Figure 1Ai), RecBCD binds to a blunt dsDNA end with K_obs = 7.0±0.8 × 10⁶ M⁻¹ with ΔH_obs =+4.6±0.1 kcal/mol and T∆S⁰ (13.9±0.1 kcal/mol). Hence, RecBCD binding to a blunt DNA end is still enthalpically unfavorable and driven by favorable entropic changes, although the ∆H_obs is substantially less unfavorable than at 50 mM NaCl. The value of K_obs decreases by at least a factor of ≥100 upon raising the [NaCl] from 50 mM to 275 mM.

One concern for the DNA binding experiments in the presence of Mg²⁺ is that the nuclease activity of RecBCD is active and could partially degrade the DNA[9, 10]. For this reason, we compared the behavior of wt RecBCD with RecB^D1080ACD, containing a single mutation within the nuclease active site that eliminates nuclease activity[10]. Our ITC experiments (Figure 2B and 2C) showed that in Buffer M50–10, RecB^D1080ACD and RecBCD bind to blunt ended DNA with the same ∆H_obs and stoichiometry. We also tested RecBCD and RecB^D1080ACD at higher [NaCl] (Buffer M275–10), where the binding affinities can be accurately measured (Figure 2D). Under these conditions, RecBCD binding to blunt ended DNA is also associated with an unfavorable ΔH_obs= +4.6±0.2 kcal/mol and driven by a favorable ∆S⁰ (T∆S⁰ = +13.9±0.1 kcal/mol). RecB^D1080ACD binds to blunt DNA ends with the same thermodynamic profiles (ΔG⁰, ΔH and TΔS⁰) as RecBCD under these conditions.

RecBCD binding to partially melted DNA ends

To determine how long the ssDNA flanking regions must be to eliminate contributions from base pair melting we examined RecBCD binding to DNA ends possessing varying lengths of twin 3’- and 5’-(dT)_n tails. As mentioned above, RecBCD shows no nuclease activity upon binding a blunt DNA end. However, as the (dT)_n tails increase in length, nuclease activity becomes likely. We compared RecBCD and RecB^D1080ACD binding to various dT_ndT_n substrates in Buffer M275–10 (Figure 2F and Supplementary Figure 1A and 1B). We found that RecBCD and RecB^D1080ACD bind with identical energetics to DNA ends possessing twin 3’-(dT)_n and 5’-(dT)_n tails with lengths up to n=15. However, we found evidence of RecBCD nuclease activity in the presence of Mg2+ with twin-tailed dT_n/dT_n DNA substrates with n≥20. Hence, RecB^D1080ACD was used to examine binding to those DNA ends.

Figure 2F shows a plot of ΔH_obs vs tail length for the dT_n/dT_n substrates. The ∆H is positive for binding to a blunt end (+4.6 ± 0.2 kcal/mol), but decreases nearly linearly with increasing tail length reaching a plateau of ∆H = −75 ±2 kcal/mol for n≥20 (for example, n=25 Figure 2E). Extrapolation of the linear region to the plateau yields an intersection at n=17–18 nucleotides. This indicates that RecBCD interacts with at least 17–18 nucleotides of a pre-melted twin tailed DNA end. The difference in ∆H between binding to a blunt DNA end and a dT_n/dT_n end with n≥25 is ∆∆H_obs = 81 ± 2 kcal/mol. Applying Eq. (3b), which assumes the end state for RecBCD bound to a blunt end is the same as RecBCD bound to the pre-melted DNA, we assign ∆H_melt = ∆∆H_obs = 81 ± 2 kcal/mol. Based on an average ∆H = +8± 1 kcal/mol bp melted[33], these data suggest that RecBCD is capable of melting significantly more than the 5–6 bp observed in RecBCD-DNA crystal structures[21, 23], possibly as many as 10–18 bp. However, the binding affinities of RecBCD and RecB^D1080ACD for dT_n/dT_n DNA (except blunt end where n=0) are too high to be measured by ITC (Supplementary Figure 1A and 1B). Thus we were unable to determine ΔG⁰ and the TΔS⁰ component, but only ∆H for these experiments even at 275mM NaCl.

RecBCD is capable of melting at least 11bp upon binding to a blunt dsDNA end

To complement our thermodynamic studies, we determined structures of RecBCD alone and in complex with the 60 bp DNA duplex used in the ITC studies. Initial cryo-EM experiments were conducted in 50 mM MOPS, pH7.0, 50 mM NaCl, 10 mM MgCl₂, but these resulted in an insufficient number of orientations of RecBCD molecules, preventing a full 3D reconstruction from the cryo-EM density maps. We then tried conditions that are similar to those used by Wilkinson et al.[24], in a previous cryo-EM study of RecBCD-DNA. We used 20 mM Tris pH 7.4, 50 mM NaCl and 4 mM MgCl₂, with amphipol added to the sample immediately before vitrification to a final concentration of either 0.0125% or 0.025%, for both RecBCD and RecBCD-DNA samples. Ultimately, the datasets for RecBCD with 0.025% amphipol and RecBCD-DNA with 0.0125% amphipol were used for data acquisition and analysis due to better quality of the grid preparations.

For cryo-EM experiments, the same 60 bp blunt-ended duplex DNA as used in the ITC studies (Supplementary Table 1) was mixed with RecBCD at a molar ratio of 1.5 DNA:RecBCD (a molar ratio of 3 DNA ends per RecBCD). Although the DNA sequences of the two ends differ slightly (7 vs. 5 GC base pairs in the first 10 base pairs), no preference was observed for RecBCD binding to one DNA end over the other over a range of [NaCl] and [MgCl₂] conditions (Figure 2 and previous studies[22, 26, 34]). The conditions used in the cryo-EM study (20 mM Tris pH 7.4, 50 mM NaCl, 4 mM MgCl₂) differ slightly from those used in most of our ITC studies (20 mM MOPS pH 7.0, 50 mM NaCl,10 mM MgCl₂ or 275 mM NaCl,10 mM MgCl₂). Under the cryo-EM solution conditions, RecBCD and RecB^D1080ACD bind stoichiometrically to each end of the 60 bp DNA with ∆H = +13.7±0.2 kcal/mol (Supplementary Figure 1Ci), slightly larger than in Buffer M275–10. We also performed ITC experiments with RecB^D1080ACD binding to DNA possessing twin dT₂₀dT₂₀ ends and dT₃₀dT₃₀ ends (Figure 2F and Supplementary Figure 1C) under the cryo-EM conditions and measured ΔH values that are only slightly less favorable than those in Buffer M275–10 (Figure 2F and Supplementary Table 2). However, the ∆∆H for binding to blunt-ended DNA vs. the pre-melted DNA ends remains the same within error as that in Buffer M275–10, with ΔΔH=81±3kcal/mol (Figure 2F).

A data set of 1468 movies were recorded and a particle picking algorithm (gautomatch) was used to extract images of RecBCD. 2D and subsequently 3D classifications were made using relion3 [49] leading to the identification of a total of 71,467 RecBCD-DNA particles showing 2 structural classes (Supplementary Figure 2A and Material and Methods). The major class, class 1, with 45,202 particles (63.2% of all particles), was globally refined (Supplementary Figure 3) to a resolution of 3.6 Å (FSC=1.43, Supplementary Figure 3). The minor class, Class 2 from 26,265 particles (36.8% of total) was globally refined (Supplementary Figure 3) to a resolution of 4.5 Å (FSC=0.143, Supplementary Figure 3). We did not observe any structural class for DNA-free RecBCD, consistent with our expectation that all RecBCD is DNA-bound under the solution conditions and concentrations used. Local resolution estimates are consistent with the global resolution estimates and show higher resolution for the majority of Class 1 compared to Class 2. (Supplementary Figure 3). Due to the weak density observed at RecD, we also performed focused 3D classification using a soft-edged mask around the RecD region. However, no further classes were identified.

The class 1 density map shows all three subunits and domains of RecBCD (Figure 3A). In addition, we observe density corresponding to an 11 nucleotide 5’ ssDNA tail as the result of DNA melting (Figure 3B). Within the RecBCD complex, the 11 nucleotides of 5’ ssDNA spans almost the entire RecD subunit, interacting with both the 1A and 2A sub-domains of the RecD motor (Figure 3C and D). However, we only observe density corresponding to 4 nt for the 3’ ssDNA end in the class 1 structure (Figure 3B). The 3’ ssDNA density spans the 2A sub-domain of the RecB motor (Figure 3E). Since the dsDNA was blunt ended, there must be at least 7 nt of 3’ ssDNA that are not resolved. The lack of expected density corresponding to a longer 3’ ssDNA tail suggests that the 3’ ssDNA between 4–11 nt is flexible. Our ITC experiments indicated (Figure 2 and Supplementary Figure 1) no nuclease activity for RecBCD binding to blunt-ended DNA even in the presence of Mg²⁺, thus eliminating the possibility that RecB^Nuc has degraded the 3’ ended DNA. In fact, flexibility of the 3’ ssDNA end was also evident in a recent cryo-EM structure examining a complex of RecBCD with dsDNA containing a Chi sequence within a long 20 nt pre-formed 3’ ssDNA region[35], such that a substantial portion of the molecules showed only weak 3’ ssDNA density beyond 4 nt. One possibility is that the 3’ ssDNA exits the protein at the RecB-RecC interface as previously suggested[34].

Figure 3. RecBCD can melt 11 bp upon binding a blunt DNA end.

(A) Cryo-EM map of the major class (class 1) of RecBCD-DNA complex, reconstructed from 45,202 particles (63.2% of total), in 20 mM Tris, pH 7.4, 50 mM NaCl and 4 mM MgCl₂, at 3.6 Å resolution. RecB motor domains (red), RecB nuclease domain (magenta), RecC (blue), RecD (green) and DNA (yellow). RecB Linker region is indicated by white dashed lines and labeled as RecB^Linker. (B) cryo-EM density shown as semi-transparent volume overlaid with cartoon representation of the DNA. This indicates that at least 11 bp can be melted upon RecBCD binding to a blunt DNA end. (C) side view of RecD (depicted as pipes and planks) highlighting the position of the 5’ ssDNA tail (yellow tube) as a result of DNA melting. RecC (blue), RecB (red), RecB^Nuc (magenta) and DNA (yellow) are colored as in panel (A). 1A, 2A, 1B and 2B subdomains of RecD are represented as pipes and planks cartoons and colored as indicated in the figure. (D) top down view of panel (C). (E) side view of RecB motor domains (colored as labeled, using the same coloring scheme as in Panel C and D) indicating that the section of 3’ssDNA that could be modeled reaches only as far as the 2A subdomain of RecB. The rest of the 3’ssDNA appears to be flexible.

Due to the fact that the DNA substrate used in our study (Supplementary Table 1) has slightly different sequences at each end, the sequence of the DNA in the cryo-EM structure could not be identified. Hence we can only conclude that a minimum of 11 bp are melted in this complex. This extent of duplex DNA melting is much larger than the 4 bp that was reported as melted in a crystal structure of a RecBCD-DNA complex[21]. For that crystal structure, Singleton et al.[21] used a 19 bp DNA hairpin with a blunt DNA end. In fact, using a slightly longer 21 bp duplex, Saikrishnan et al.[23] observed melting of 6 bp from a dsDNA end that also possessed a 5’ dT₄ overhang. Saikrishnan et al.[23] hypothesized that the shorter 19 bp duplex DNA might affect the binding of RecBCD, and limit the amount of dsDNA that is melted. However, it is possible that a 21 bp duplex is also too short to allow the full DNA melting that we observe with the 60 bp DNA. Interestingly, only 15 bp of duplex DNA can be resolved in both the class1 and class 2 RecBCD-DNA structures indicating that the remaining duplex region must be too flexible to be resolved. This is the same length of duplex DNA that was resolved in the two crystal structures[21, 23] and a previous cryo-EM structure[36] of RecBCD-DNA complexes. This appears to be the duplex DNA length that is needed to fully interact with the arm domain of RecB. This suggests that the shorter DNA lengths used in previous structural studies of RecBCD-DNA binding might limit the extent of DNA melting. The 60 bp duplex DNA used in our studies likely possesses no such limitation. This 11 bp melted region is consistent with our thermodynamic binding studies (Figure 2) that suggested that RecBCD is capable of melting 10–18 bp from a blunt ended duplex DNA.

RecBCD-DNA complex shows significant heterogeneity in both DNA melting and conformations

In contrast to the major class 1 structure (Figure 3), a minor class 2 RecBCD-DNA structure (36.8% of particles), reconstructed from 26,265 particles at 4.5 Å resolution, shows weak density for RecD and complete lack of density for RecB^Nuc (Figure 4A). Since we performed denaturing gel electrophoresis and analytical sedimentation velocity experiments on the RecBCD used in the cryo-EM samples, we are confident that it contain full length subunits with no partial degradation. Our thermodynamic binding studies also corroborate with this. RecBCD has significantly higher binding affinity to DNA ends than does RecBC. We have also found that a nuclease deletion variant, RecB^ΔNucCD, has even higher DNA binding affinity than RecBCD[37] (Hao and Lohman, unpublished). Hence, if our RecBCD contained significant amounts of RecBC or RecB^ΔNucCD, this would be detectable from the binding isotherms (Figure 2 and Supplementary Figure 1). Furthermore, the class 2 structure shows strong density corresponding to the N-terminal domain of RecD (Figure 4B) indicating that the weak density of RecD in the class 2 structure is due to conformational flexibility rather than averaging of RecBCD and RecBC. On the other hand, the density for RecB^Nuc/RecB^Linker is fully evident in the class 1 structure (Figure 3A), but not observed at all in the class 2 structure (Figure 4A and 4B). This suggests that the entire RecB^Nuc/B^Linker is either docked in the position shown in Figure 4A or undocked and dynamic, adopting an ensemble of conformations. Such conformational flexibility of the RecB^Nuc domain was suggested in earlier studies[18, 38] and is consistent with the different structural classes evident in the current study. However, we did not observe any RecBCD-DNA structures that showed an alternative stable docking site of the RecB^Nuc that differs from its position in the class 1 structure as has been suggested[38].

Figure 4. Heterogeneity of RecBCD-DNA complex.

(A) Cryo-EM map of the minor class of RecBCD-DNA complex (class 2), reconstructed from 26,265 particles (36.8% of total), in 20 mM Tris, pH 7.4, 50 mM NaCl and 4 mM MgCl₂, at 4.5 Å resolution. RecB motor domains (red), RecB nuclease domain (magenta), RecC (blue), RecD (green) and DNA (yellow). RecD density is weak. (B) Back view of panel A (180 rotation along y axis), highlighting RecD N terminal domain in pink. (C) cryo-EM density shown as semi-transparent volume overlaid with cartoon representation of the DNA corresponding to 4 nucleotides on both the 3’ and 5’ strands are observed. (D) cutaway of RecC showing that the 5’ ssDNA tail does not reach the RecD subunit. Molecular model could be built only for the RecD N terminal domain (depicted as pink pipes) due to weak density for RecD.

In addition, for the class 2 RecBCD-DNA structure, we only observe ssDNA density corresponding to 4 nt for both the 3’ and 5’ ends (Figure 4C). A 5’ ssDNA tail of 4 nucleotides is embedded in the RecC channel, but is not long enough to interact with RecD (Figure 4D). However, there are no differences in the DNA contacts between class 1 and class 2 RecBCD-DNA structures in the regions that could be resolved. The differences in DNA densities for the class 1 vs. class 2 structures could be a result of the poorly ordered RecD subunit in the class 2 structure leading to flexibility in the bound ssDNA. As mentioned above, since we are unable to identify the DNA sequence in these structures, we can only conclude that a minimum of 4 bp are melted in the class 2 structure, hence it is possible that both class 1 and class 2 structures have the same amount of DNA melted. However, it is also possible that the class 2 RecBCD-DNA complex has melted only 4 bp of the DNA duplex. The latter becomes a much more likely explanation when we compare the structures of RecBCD (Figure 5) with RecBCD-DNA (Figure 3 and 4) as discussed below. Our results suggest that the extent of DNA melting by RecBCD is not uniform within the ensemble of RecBCD-DNA complexes. In fact, heterogeneity within the ensemble of RecBCD-DNA complexes has been reported previously[18, 27, 39]. The interaction of RecD with the DNA and/or the docking of the RecB^Nuc domain could be responsible for the heterogeneity in DNA melting as discussed below.

Figure 5. Conformational heterogeneity of RecBCD.

(A), (B) and (C) show cryo-EM density maps of RecBCD alone reconstructed from class 1 (61,888 particles, 3.7 Å resolution), class 2 (30,267 particles, 4.2 Å resolution) and class 3 (48,007 particles at 4.3 Å resolution) in 20 mM Tris, pH 7.4, 50 mM NaCl and 4 mM MgCl₂. Segments of the maps (RecB (red), RecB^Nuc (magenta), RecC (blue) RecD (green)) as labeled in (A). RecB Linker region is indicated by white dashed lines and labeled as RecB^Linker in (A). In (A) and (C) the majority of the density corresponding to RecD was not observed, except for the N terminal domain. Part of the RecD density corresponding to its 2A and 2B subdomains in class 2 (B) is also substantially weaker compared to the rest of the RecD subunit and could not be used to build a molecular model. Density corresponding to RecB^Nuc and RecB^Linker was evident in class 1 (A), but completely missing for both class 2 (B) and class 3 (C) indicating that RecB^Nuc is undocked from RecC in those classes.

RecBCD shows greater conformational heterogeneity than RecBCD-DNA complexes

Interestingly, all reported structures of RecBCD to date have been in complex with DNA. To assess the effect of DNA binding, we examined the RecBCD hetero-trimer in the absence of DNA. These experiments were performed in 20mM Tris, pH 7.4, 50mM NaCl and 4mM MgCl₂, with a final concentration of 0.025% amphipol that was added immediately before vitrification. 2,353 movies were recorded and a similar workflow was applied for data analysis (Supplementary Figure 2B). 2D and subsequently 3D classifications were made using relion3 [49] leading to the identification of a total of 14,016 RecBCD particles representing three structural classes (Supplementary Figure 2B and Material and Methods). 3D cryo-EM density maps were reconstructed for these three structural classes through global refinement using relion3 [49] (Supplementary Figure 2B) at resolutions of 3.7Å for class 1 (61,888 particles in the final map), 4.2 Å for class 2 (30,267 particles in the final map) and 4.3Å for class 3 (48,007 particles in the final map) (Supplementary Figure 3). Estimates of local resolution (Supplementary Figure 3) show high resolution of the core of the protein complex in each class but lower resolution especially for parts of RecB and RecD.

The most striking differences among the three classes of RecBCD structures are found in the densities corresponding to RecB^Nuc, the RecD subunit and the ~70 residue linker, RecB^Linker, connecting the RecB^Nuc to the RecB motor (Figure 5). In class 1, densities for RecB^Nuc, and RecB^Linker are clearly evident. However, only weak density is observed for most of the RecD subunit (Figure 5A). In class 2, densities for the RecB^Nuc and RecB^Linker are not observed, although stronger map density is evident for RecD than in class 1, however only weak density is observed for the half of RecD corresponding to RecD^2A and RecD^2B (Figure 5B). In class 3 (Figure 5C), densities are not observed for both RecB^Nuc/B^Linker and the majority of RecD. As reasoned earlier, the weak density corresponding to RecD indicates conformational flexibility rather than averaging between RecBCD and RecBC as a result of some RecD dissociation. On the other hand, the lack of RecB^Nuc density suggests that the nuclease domain is not docked at a unique position, but is undocked and dynamic. Besides these major conformational differences, we also observed conformational shifts in other domains of the protein complex but at much smaller scales among the three RecBCD classes.

In stark contrast to the RecBCD-DNA structures, where the majority of particles (63.2%) show a docked RecB^Nuc domain and strong RecD density, all of the RecBCD structures show weak RecD density and only 44.2% display a docked RecB^Nuc domain (Figure 5). This suggests that DNA binding promotes docking of RecB^Nuc to RecC and reduces the conformational heterogeneity of RecD. However, even with DNA bound, a significant population (37%) of RecBCD-DNA complexes (class 2) still shows weak RecD density and no density for RecB^Nuc. While our data cannot rule out the possibility that class 1 and class 2 RecBCD-DNA complexes result in the same amount of DNA melting, given the stabilizing effect of DNA binding on RecBCD conformations, it is more likely that class 2 RecBCD-DNA complexes lack the RecD-DNA interactions evident in the class 1 complexes. These observations add support to our suggestion that the class 2 RecBCD-DNA complexes likely melt significantly less DNA (4 bp) than the class 1 complexes. Thus, in the class 2 structures, the shorter 5’ ssDNA tail that results from less DNA melting is unable to reach and stabilize RecD.

Our results indicate that conformational heterogeneity is evident in both RecBCD and RecBCD-DNA complexes and that the extent of DNA melting within an ensemble of RecBCD-DNA complexes is not uniform. It is important to note that the classes of RecBCD and RecBCD-DNA structures observed in the cryo-EM data sets represent the most populated species and that they can represent an averaging of species with different conformations that are in dynamic equilibrium in solution.

Discussion

RecBCD is able to melt at least 11 bp upon binding to a blunt DNA end

SF1 helicases generally require a ssDNA loading site to initiate DNA unwinding. However, using only its binding free energy, RecBCD is able to initiate DNA unwinding on a fully base paired DNA end by melting a region of the DNA end to form the ssDNA needed to engage the ssDNA translocation motors. Farah and Smith[20] first reported that ~4–5 bp can be disrupted or melted upon RecBCD binding to a blunt DNA end in a Mg²⁺-dependent, but ATP-independent reaction. This was later supported by the first crystal structure of RecBCD bound to a DNA end[21] showing that 4 bp of a blunt DNA end were indeed melted in a complex formed in the presence of Ca²⁺. Wong et al.[26] and Wong and Lohman[22] later concluded from quantitative DNA binding studies similar to those used in the current study, as well as footprinting studies, that RecBC is also able to melt 6 bp upon binding to a blunt DNA end. Hence it has been generally viewed that RecBCD can melt 6 bp upon binding to a blunt DNA end using only its binding free energy in a reaction that requires a divalent cation. Consistent with this, we also have shown that RecBCD binding to a blunt DNA end is enhanced by Mg²⁺ and accompanied by the uptake of one Mg²⁺ in 30 mM NaCl (Hao and Lohman unpublished). However, the thermodynamic and structural studies reported here show that RecBCD can bind to a blunt DNA end and melt more than 11 bp and possibly up to 17–18bp in the absence of ATP.

Our ITC experiments show that RecBCD binding to a blunt DNA end is entropically driven and associated with an unfavorable positive ∆H (+4.6±0.2 kcal/mol at 275mM NaCl (Buffer M275–10) and +15.5±0.2kcal/mol at 50mM NaCl (Buffer M50–10), 25.0°C). This unfavorable ∆H results in part from the fact that RecBCD binding melts some DNA base pairs and that base pair melting is associated with a large unfavorable ΔH[33, 40, 41]. SantaLucia[33] reported a range of ∆H for base pair melting from +7.2 to +10.6 kcal/mol bp depending on the DNA sequence, with an average ∆H = +(8± 1) kcal/mol bp. Other estimates range from +4.3 to +9 kcal/mol bp[41], and +5.2 to +15 kcal/mol bp[40] depending on base composition and temperature. If 11 bp are melted as we observe in the class 1 RecBCD-DNA structures, this would result in an unfavorable contribution to the overall ∆H of +88 kcal/mol, using the average of +8 kcal/mol bp. Hence, there must be large favorable contributions to the overall ∆H due to RecBCD-DNA interactions that offset most of the unfavorable ∆H due to bp melting. Our ITC studies of RecBCD binding to a fully pre-melted DNA end provide an estimate for this compensating favorable ∆H since the contributions from bp melting are eliminated. At 275 mM NaCl (Buffer M275–10, 25°C) the binding enthalpy starts from ∆H= +4.6±0.2 kcal/mol for a blunt DNA end, but decreases nearly linearly with increasing length of twin pre-melted dT_n/dT_n DNA ends, reaching a plateau of ∆H= −76±2 kcal/mol for n≥20. Extrapolation of the linear region to the plateau yields an intercept at n=17.5 nucleotides. The difference in ∆H for RecBCD binding to a blunt dsDNA end vs a fully melted DNA end is ∆∆H= +(81±2) kcal/mol (Buffer M275–10, 25°C). The same ITC experiments performed in the cryo-EM buffer yield a similar ∆∆H = +81±3 kcal/mol. These ∆∆H values provide an estimate of the unfavorable ∆H contributed by bp melting, a large part of it being compensated by favorable protein-DNA interactions in the melted DNA complex. This ignores the potential energetic differences for RecBCD binding to the dT₂₀ tails vs. dN₂₀ tails of mixed base sequence.

In similar ITC experiments, Wong and Lohman[22] reported ΔH for RecBC binding to the same series of dT_n/dT_n DNA substrates (in Buffer M100–10, 25°C). RecBC showed a favorable ∆H = −17±4 kcal/mol for binding to a blunt DNA end, which becomes more favorable in a nearly linear dependence for DNA ends with longer twin dT_n/dT_n tails, reaching a plateau at ∆H = −64±3 for n≥ 6, up to n=20, yielding ∆∆H=47±7 kcal/mol for RecBC. This difference was attributed to the cost of melting 6 DNA bp from a blunt DNA end at 25°C, amounting to an average ∆H = +8±1 kcal/mol of bp melted, nearly identical to the average determined from DNA melting studies[33, 40, 41]. However, the observation of a favorable ∆H_obs = −17±4 kcal/mol for RecBC binding to a blunt DNA end suggests that the unfavorable ∆H_obs = +47±7 kcal/mol for the melting of 6 bp is more than compensated by the favorable interactions within the RecBC-DNA melted complex.

The ∆∆H= 81±2 kcal/mol for RecBCD is much greater than that of RecBC (∆∆H=47±7 kcal/mol), indicating that RecBCD has the capability of melting more DNA base pairs than RecBC. Using ∆H = +8±1 kcal/mol per DNA bp melted, our current studies suggest that RecBCD can melt (81±2)/(8±1) = 9–12 bp from a blunt DNA end, similar to what we observe in the class 1 RecBCD-DNA structure. However, the observation that a plateau in ∆H for RecBCD binding to the pre-melted DNA ends is not reached until n= 17–18 indicates that RecBCD can make favorable contacts with ssDNA regions up to 18 nt long, although there is an enthalpy/entropy compensation keeping the ∆G° nearly constant for n≥10. This suggests that RecBCD may be capable of melting up to 18 bp, with an enthalpic cost of ∆H =+4.5±0.3 kcal/mol per DNA bp melted, still within the range of previous estimates from DNA melting[33, 40, 41].

The cryo-EM class 1 structures presented here also show that RecBCD can melt at least 11 bp upon binding to a blunt DNA end. Although the first RecBCD-DNA crystal structure[21] showed only 4 bp melted from a blunt DNA end, this may be due to differences in the DNA substrates and conditions used in those studies. In the crystallographic study, Singleton et al.[21] used a 19 bp duplex DNA in buffers with 100 mM NaCl and Ca(AcO)₂, while our cryo-EM studies used a 60 bp duplex in buffer with 50 mM NaCl and 4 mM MgCl₂. Saikrishnan et al.[23] suggested that the observation that only 4 bp were melted in the first crystal structure[21] was due to a limitation imposed by the short DNA duplex length of 19 bp, since they observed 6 bp melted using a longer 21 bp duplex[23]. However, our results suggest that even the 21 bp duplex may have reduced bp melting. Our use of a 60 bp duplex may have removed this limitation.

In the cryo-EM structure reported by Wilkinson et al.[24], the DNA end bound to RecBCD possessed a forked structure with a 3 nucleotide unpaired 3’-ssDNA and a 12 nucleotide unpaired 5’-ssDNA that spans the entire RecD subunit. This 5’-ssDNA tail is similar in length to the 11 nucleotide 5’-ssDNA observed in the class 1 RecBCD-DNA structure reported here. The DNA contacts observed in the RecBCD-DNA class 1 structure are the same as those in the Wilkinson et al.[24] structure. The conformations of the backbones and the orientations of the ssDNA bases are very similar. In fact the RMSD between these structures is only 0.7Å.

Conformational heterogeneity of RecBCD and RecBCD-DNA

Several published structures of RecBCD in complex with various DNA ends have been reported[21, 23, 24, 35]. Here we report the first structures of DNA-free RecBCD. While our ensemble experiments demonstrate that our RecBCD sample is a chemically monodisperse hetero-trimer in solution, our cryo-EM structures indicate that RecBCD displays significant conformational heterogeneity, suggesting a dynamic enzyme. In particular, the RecD subunit, RecB^Nuc and RecB^Linker all show large conformational variabilities among the three RecBCD structural classes identified. The density corresponding to RecB^Linker and RecB^Nuc suggests that RecB^Nuc is docked on RecC, as observed in the crystal structures[21], in only 44% of the RecBCD particles (RecBCD class 1, Figure 5A). However, we did not observe RecB^Nuc docked at any alternative sites indicating that for the rest of the particles (56%), the RecB^Linker and RecB^Nuc, while still connected to the RecB motor domain, are dynamic in solution (Figure 5B and 5C).

Interestingly, binding of RecBCD to a blunt DNA end significantly reduces these conformational heterogeneities since we identified only two classes of RecBCD-DNA structures (Figure 3A and 4A). The major class 1 structure (63.2% of the particles) showed strong density for all subunits and structural domains and is similar to the first reported RecBCD-DNA crystal structure[21]. The entirety of the RecD subunit and the RecB^Linker are well ordered, and the RecB^Nuc is docked onto RecC. For the remaining 36.8% of particles in the class 2 structures, large parts of the RecD subunit showed weak density and with no density evident for the RecB^Nuc/B^Linker regions. This suggests that RecBCD-DNA complexes are still conformationally heterogeneous and dynamic.

In addition, the class 2 RecBCD-DNA structure (36.8%) shows density corresponding to only 4 nt of ssDNA on both the 5’ and 3’ ended strands. Although it is possible that there is additional ssDNA in this structure that is not resolved, it is likely that less DNA is melted in this structural class possibly due to RecD not interacting with the 5’ ssDNA tail (Figure 4A). We have shown before that compared to a DNA end possessing 3’dT₆ and 5’dT₁₀ tails, RecBCD requires additional kinetic steps to initiate dsDNA unwinding from a blunt DNA end or an end possessing only 3’dT₆ and 5’dT₆ tails[29]. 5’dT₆ is not long enough to reach the RecD subunit. This suggests the importance of RecD-DNA interactions in the initiation of dsDNA unwinding. Single-molecule studies have also observed ATP-independent conformational dynamics when RecD interacts with a long enough 5’ ssDNA tail[30, 31]. It is possible that the class 2 structure represents a complex that is not able to immediately initiate DNA unwinding upon addition of ATP (Figure 6). In fact, Lucius et al.[28] showed that only ~80% of the blunt ended DNA molecules are unwound rapidly by RecBCD in an ensemble stopped-flow experiment even though all of the DNA is bound by RecBCD. Our observation of two classes of RecBCD-DNA complexes may explain this observation if only one structural class (presumably class 1) is active for DNA unwinding (Figure 6). We suggest the class 2 structure must first transition to the class 1 structure before DNA unwinding can proceed upon ATP binding. It is important to emphasize that the classes of RecBCD and RecBCD-DNA structures we identified here are snapshots of the most populated states within an ensemble and that in solution, these states are in dynamic exchange. Single-molecule studies[30, 31] also suggest that RecBCD can transiently melt a DNA duplex. The states of RecD and RecB^Nuc can also change depending on how much ssDNA interacts with RecBCD (Figure 6). However, our results suggest that interactions with long enough ssDNA tails can stabilize the RecBCD complex and favor a state that is active for DNA unwinding (Figure 6).

Figure 6. Cartoon model summarizing the RecBCD conformational and DNA melting heterogeneities.

In the absence of DNA, RecBCD shows conformational heterogeneity with only 44% of the particles showing the RecB^Nuc docked at RecC and the RecD subunit is conformationally flexible. When undocked from RecC, RecB^Nuc is dynamic in solution while remaining tethered to the RecB motor domain by the RecB^Linker. Significant heterogeneity is also observed for RecBCD when bound to a blunt DNA end. The majority of RecBCD (63.2%) shows melting of at least 11 bp of DNA along with a stabilized RecD and docked RecB^Nuc. This class of RecBCD-DNA complex may represent a complex that is active for DNA unwinding. This complex is in equilibrium with a smaller but still substantial population of RecBCD-DNA complex that shows melting of only 4 bp of DNA and shows a more flexible RecD subunit and an undocked RecB^Nuc and may represent an unproductive initiation complex. These likely represent complexes in which the extent of DNA melting transitions dynamically between 4–11 bp. The different classes of RecBCD and RecBCD-DNA complex are also in equilibrium with DNA binding favoring a docked RecB^Nuc domain.

RecBCD also facilitates the loading of RecA protein onto the 3’ resected DNA strand in order to initiate recombinational repair of a double-stranded DNA break. Spies et al.[19] showed that the RecB^Nuc domain interacts with RecA protein and is likely involved in RecA loading. Furthermore, the region of RecB^Nuc that interacts with RecA is occluded through interactions with RecC in the RecBCD-DNA crystal structure[19], as well as in the class 1 structure reported here. This suggests that the RecB^Nuc domain must move from its docked position in order for it to bind RecA protein although evidence for this has been lacking. The different classes of RecBCD and RecBCD-DNA structures that we present here provide direct evidence that RecB^Nuc can exist in both a docked and undocked state.

Interestingly, the structural heterogeneity that we report here may provide an alternate explanation for the DNA unwinding heterogeneity reported by Liu et al.[39] based on their single-molecule studies. Liu et al.[39] observed that the DNA unwinding rates of individual RecBCD molecules varied considerably, although the rate of unwinding by an individual RecBCD remained constant. However, a transient pause of DNA unwinding by depleting any RecBCD of ATP generally resulted in a change in the DNA unwinding rate upon resumption of DNA unwinding. The fraction of RecBCD enzymes that changed rates after the pause increased with the duration of the pause time. Liu et al.[39] interpreted this result as evidence that RecBCD exists in multiple conformational sub-states that can equilibrate on a time scale of seconds in the absence of ATP and that each sub-state determines the rate of DNA unwinding. Liu et al.[39] hypothesized that the source of the DNA unwinding heterogeneity might reflect whether only one or both of the translocation motors (RecB and RecD) are engaged with the DNA since mutation of either the RecB or RecD motors results in slower unwinding by the mutant RecBCD[6–8, 42]. Our structural studies suggest that this DNA unwinding heterogeneity could also result from whether the RecB nuclease domain is docked onto RecC since RecB^Nuc docked structures also appear to stabilize the RecD motor. In further support of this proposal, we have shown that the rate of DNA unwinding decreases for a RecBCD variant that is missing the RecB^Nuc domain[13].

Material and Methods

Buffers

Reagent grade chemicals and double-distilled water further deionized with a Milli-Q purification system (Millipore Corp., Bedford, MA) were used to make all buffers in this study. Buffer A contains 50 mM Tris HCl, pH 7.5, 10% sucrose. Buffer C is 20 mM potassium phosphate, pH 6.8, 0.1 mM β-mercaptoethanol, 0.1mM EDTA, 10% (v/v) glycerol. Buffer M contains 20 mM MOPS-NaOH, pH 7.0, 1mM β-mercaptoethanol, 5% (v/v) glycerol. We determined the concentration of stock MgCl₂ solutions by measuring their refractive indexes using a Mark II refractometer (Leica Inc., Buffalo, NY). All DNA binding experiments were performed in Buffer M or cryo-EM Buffer (20 mM Tris pH 7.4, 50 mM NaCl, 4 mM MgCl₂). Our buffer nomenclature is Buffer MX-Y and TX-Y, where X indicates the [NaCl] and Y indicates the [MgCl₂] in mM concentration units (e.g., Buffer M30–10 has 30 mM NaCl and 10 mM MgCl₂).

Proteins and DNA

RecBCD hetero-trimer was purified as described[13, 27, 28]with modifications[37]. A Hitrap heparin column was used to separate RecBCD hetero-trimer from hetero-hexamer ((RecBCD)₂)[13, 32]. However, a significant portion of RecBCD remains as hetero-hexamer form. We found that some of the RecBCD hetero-hexamer can be converted to hetero-trimer after an ~8 hr incubation in Buffer C + 2M NH₄Cl, followed by dialysis vs. Buffer C. Any remaining hetero-hexamer was separated from hetero-trimer by repeating the Hitrap Heparin column. Purified RecBCD was dialyzed into Buffer C, aliquoted and flash-frozen in liquid nitrogen and stored at −80°C. RecBCD concentration was determined by absorbance in Buffer C, using an extinction coefficient[27] of ε₂₈₀=4.5×10⁵ M⁻¹ cm⁻¹. Bovine serum albumin (BSA, from Sigma St. Louis, MO) concentration was determined by absorbance spectrum using an extinction coefficient of ε₂₈₀=4.3×10⁴M⁻¹cm⁻¹ in Buffer C[26].

Oligodeoxynucleotides were synthesized using a MerMade 4 synthesizer (Bioautomation, Plano, TX) with phosphonamidite reagents (Glen Research, Sterling, VA) and purified as described[43]. Concentrations of each oligodeoxynucleotide were determined spectrophotometrically as described[13, 40]. Double stranded DNA was formed by annealing the two complementary single stranded oligodeoxynucleotides by heating the mixture to 95°C for 5 minutes followed by slow cooling to 25°C.

Sedimentation Velocity

Sedimentation velocity experiments were performed at 42000 rpm, 25°C, using an An50Ti rotor in an Optima XL-A analytical ultracentrifuge (Beckman Coulter, Fullerton, CA, USA). The concentrations of RecBCD used were between 0.3–1 µM, and sedimentation was monitored by absorbance at 280 nm. SEDNTERP [44] was used to determine the density and viscosity of the buffers at 20°C and the partial specific volume of RecBCD (0.736ml/g) in Buffer M50–10. Sedimentation data were analyzed using SEDFIT to yield continuous sedimentation coefficient distributions, c(s)[44, 45]. The sedimentation coefficients were converted to s_20,w using SEDFIT[44, 45] and plotted in Figure 1B.

Isothermal Titration Calorimetry

ITC experiments were performed using a VP-ITC calorimeter (Malvern Panalytical, Malvern, UK) as described[26, 46]. Solutions of RecBCD and DNA were extensively dialyzed against the reaction buffer at 4°C. Samples were then centrifuged to remove any insoluble particulates and degassed before use. A solution of RecBCD (0.5 to 1µM in the sample cell) was titrated with 10 µl injections of DNA (3–5µM in the syringe) at 5 min intervals with a stirring rate of 130 rpm. Separate control experiments were performed to determine the heat of dilution for each injection by injecting the same volumes of DNA into the sample cell containing only buffer. An N independent and identical sites model was used to analyze the total heat after the i-th injection $\left(Q_i^{tot}\right)$ as a function of [DNA] using Eq. (4) to obtain the observed enthalpy change (ΔH_obs) and equilibrium binding constant (K_obs) for RecBCD binding to each DNA end, and the binding stoichiometry (N), although N was floated, the non-linear least squares analysis always yielded N=2 within a 2% uncertainty.

$$Q_i^{tot}=V_0\mathrm{\Delta }{H}_{obs}{M}_{tot}\frac{N{K}_{obs}X}{1+K_{obs}X}$$ (4)

We emphasize that in equation (4) ΔH_obs and K_obs are the values for RecBCD binding to one end of a dsDNA substrate. V₀ is the volume of the calorimetry cell (1.43ml), M_tot is the total DNA concentration and X is the free RecBCD concentration and is obtained by solving Eq. (5),

$$X_{tot}=X+X_{bound}=X+\frac{N{K}_{obs}X}{1+K_{obs}X}{M}_{tot}$$ (5)

where X_tot is the total RecBCD concentration of in the cell after the i^th injection. When K_obs can be measured (10³M⁻¹<K_obs<10⁹M⁻¹), the (1 M) standard state binding free energy (ΔG⁰) and the entropy change of binding (TΔS⁰) are calculated from eqs. (6) and (7), respectively. All uncertainties are reported

$$\mathrm{\Delta }{G}^0=-RTln(K_{obs})$$ (6)

$$\mathrm{\Delta }{G}^0=\mathrm{\Delta }H-T\mathrm{\Delta }{S}^0$$ (7)

at the 68% confidence limit (± one standard deviation).

Cryo-EM sample preparation and imaging

For preparation of cryo-EM grids, RecBCD was extensively dialyzed vs. a buffer containing 20mM Tris, pH 7.4, 50mM NaCl and 4mM MgCl₂. RecBCD was concentrated to 10µM (determined spectrophotometrically) using Vivaspin 500 centrifugal concentrators (Sartorius Stedim Biotech, NY) followed by centrifugation to remove any insoluble material. RecBCD-DNA complexes were formed by adding the blunt-ended dsDNA substrate (in the same buffer as RecBCD) to a final concentration of 15µM and allowed to incubate on ice for at least 15 minutes. Right before grid preparation, amphipol A8–35 (Anatrace, OH) was added to a final concentration of either 0.025% or 0.0125% for both RecBCD and RecBCD-DNA samples. Ultimately, the datasets of RecBCD with 0.025% amphipol and RecBCD-DNA with 0.0125% amphipol were used for data acquisition and analysis due to better quality of the grid preparations.

Grids were prepared and imaged at the Washington University Center for Cellular Imaging (WUCCI). Immediately after addition of amphipol, 3µl of RecBCD or RecBCD-DNA solution was applied to holey carbon grids (Quantifoil R2/2 300mesh) that were glow discharged. The grids were blotted using FEI Vitrobot Mark IV (FEI) at 100% humidity for 2s and plunge-frozen into liquid ethane. The prepared grids were imaged using a Titan Krios (FEI) G3 electron microscope operating at 300kV with a Gatan K2-Summit detector (Gatan) on the end of a BioQuatum 968 GIF quantum energy filter (Gatan) using a slit width of 20 eV. Images were recorded with EPU software (ThermoFisher Scientific) with a pixel size of 1.1 Å and a nominal defocus range of −1.0 to −2.5µm. Data were collected with a dose rate of 1.65 e/Ų per frame over a total of 40 frames with a frame rate of 0.2 s/frame and a total dose of 66 e/Ų.

Image processing and model building

The image processing workflows are summarized in Supplementary Figure 2. The RecBCD and RecBCD-DNA datasets were processed similarly with similar strategies. Corrections for beam-induced motion and dose weighting were performed using MotionCorr2[47]. The contrast transfer function (CTF) was determined using GCTF[48]. Gautomatch was used for automated particle picking. Extracted particles were subjected to two rounds of two-dimensional (2D) classification using a particle box size of 250 pixels. 2D classes that exhibit high quality secondary structure features were manually selected and further processed in the second round of 2D classification. The resulting particles were used to generate a de novo three-dimensional (3D) initial model using Relion 3[49]. 3D classifications were carried out using the initial model as a reference map. For RecBCD, 3D classification produced 3 structural classes of RecBCD particles with a total of 140162 particles. Poorly aligned particles were discarded. 3D refinement and post-processing were carried out for each of the three classes, resulting in an overall resolution of 3.7Å for the major class and 4.2Å and 4.3Å for the other two (Supplementary Figure 2B and Supplementary Figure 3). For RecBCD-DNA, 3D classification produced 2 classes of high quality RecBCD-DNA particles. 3D refinement and post-processing produced maps with overall resolutions of 3.6Å and 4.5Å (Supplementary Figure 2A and Supplementary Figure 3). Relion 3[49] was used to calculate local resolutions.

For model building, the atomic model of RecBCD in complex with dsDNA from another cryo-EM study (PDB 5ld2)[24] was used as a template for both our RecBCD and RecBCD-DNA structures. It was first fit into our cryo-EM maps using UCSF Chimera[50]. An initial round of rigid body refinement was performed using PHENIX[51], followed by cycles of real_space_refine in PHENIX[51] and model building in COOT[52]. Structural figures were made using UCSF ChimeraX[53]. Model statistics (Supplementary Tables 3 and 4) were generated using PHENIX real-space refinement[51]. The coordinates of the final models and cryo-EM maps have been deposited at wwPDB and EMDB respectively.