Rad51 Interaction Analysis Reveals a Functional Interplay Among Recombination Auxiliary Factors

Although Rad51 is the key protein in homologous recombination (HR), a major DNA double-strand break repair pathway, several auxiliary factors interact with Rad51 to promote productive HR. Here, we present an interdisciplinary characterization of the interaction between Rad51 and Swi5-Sfr1, a widely conserved auxiliary factor. NMR and site-specific crosslinking experiments revealed two distinct sites within the intrinsically disordered N-terminus of Sfr1 that cooperatively bind to Rad51. Although disruption of this binding severely impaired Rad51 stimulation in vitro, interaction mutants did not show any defects in DNA repair. Unexpectedly, in the absence of the Rad51 paralogs Rad55-Rad57, which constitute another auxiliary factor complex, these interaction mutants were unable to promote DNA repair. Our findings provide molecular insights into Rad51 stimulation by Swi5-Sfr1 and suggest that, rather than functioning in an independent subpathway of HR as was previously proposed, Rad55-Rad57 facilitates the recruitment of Swi5-Sfr1 to Rad51.

present at much higher concentrations than full-length Swi5-Sfr1, these results led to a model 1 in which Sfr1N keeps Swi5-Sfr1C anchored in close proximity to Rad51 25 . 2 Due to the use of truncated proteins in which entire domains were deleted 25 , it was not 3 possible to determine whether Sfr1N has any function other than anchoring Swi5-Sfr1 to 4 Rad51. To explore this, we employed an interdisciplinary approach to further characterize 5 Sfr1N. We provide direct evidence that Sfr1N is intrinsically disordered and contains two sites 6 that interact cooperatively with Rad51. Mutation of critical residues within these two sites 7 rendered Rad51 refractory to the stimulatory effects of full-length Swi5-Sfr1, mimicking the 8 results obtained with Swi5-Sfr1C (i.e., when the N-terminus of Sfr1 is absent), indicating that 9 the sole function of Sfr1N is to facilitate the interaction between Swi5-Sfr1 with Rad51. 10 Unexpectedly, and in contrast to the severely impaired Rad51 stimulation observed in vitro, 11 these interaction mutants only showed defects in Rad51-mediated DNA repair in the absence 12 of Rad55-Rad57, implying that these Rad51 paralogs can promote the recruitment of Swi5-13 Sfr1 to Rad51. Collectively, these results provide a molecular basis for Rad51 stimulation by 14 Swi5-Sfr1 and reveal a novel interplay between recombination auxiliary factors. 15 NC formation at substoichiometric concentrations. Nevertheless, the loss of Rad51 stimulation 1 observed at higher concentrations of wild type Swi5-Sfr1 was attenuated in the 3A and 4A 2 mutants ( Fig. 5c lanes 6, 21 and 28), suggesting that this loss of stimulation occurs due to 3 unproductive interactions between Swi5-Sfr1 and Rad51. Consistent with this notion, efficient 4 stimulation of Rad51 was maintained at higher concentrations of the 7A mutant ( Fig. 5c lanes 5 8 and 15) and Swi5-Sfr1C (Fig. 4b,c) 25 . Collectively, these results indicate that interactions 6 between Rad51 and both Sites 1 and 2 are important for efficient stimulation of strand 7 exchange. 8 9

Rad51 filament stabilization and ATPase stimulation is mediated by Sites 1 and 2 10
To determine why stimulation of Rad51-driven strand exchange is inefficient when Sites 1 and 11 2 are mutated, the molecular roles of Swi5-Sfr1 were considered. At substoichiometric 12 concentrations, Swi5-Sfr1 effectively stabilizes Rad51 filaments 21 . Thus, it seemed feasible 13 that the observed impairment in strand exchange might be explained by defects in Rad51 14 filament stabilization. To test this possibility, filament stability was examined by fluorescence 15 anisotropy. When Rad51 binds to a fluorescently-labelled oligonucleotide and forms a filament, 16 the fluorescence anisotropy increases due to a retardation in the motion of the labelled 17 oligonucleotide (Fig. 6a). The dissociation of Rad51 is accompanied by a reduction in 18 anisotropy, with the rate of decline reflective of Rad51 filament stability. Rad51-ssDNA 19 filaments were formed in the presence of ATP and filament collapse was induced via dilution 20 into reaction buffer containing ATP but lacking DNA and protein. In the absence of Swi5-Sfr1, 21 the decrement in anisotropy was sharp and reached a value that was observed in the absence 22 of protein (~0.1) within ~500 seconds. Inclusion of wild type Swi5-Sfr1 resulted in a slower 23 reduction in anisotropy, indicating that the Rad51 filament had been stabilized (Fig. 6b). 24 Strikingly, inclusion of the 7A mutant did not result in any obvious filament stabilization (Fig.  25 6c). Furthermore, although both 3A and 4A mutants showed some stabilization of Rad51 26 filaments, the magnitude of this stabilization was less than that observed for wild type protein dissociation of Rad51-ssDNA complexes showed a substantial decline in the presence of 1 Swi5-Sfr1, a lesser decline for the 3A and 4A mutants, and only a marginal decline for the 7A 2 mutant (Fig. 6f). Taken together, these results indicate that Sites 1 and 2 within Sfr1N interact 3 cooperatively with Rad51 to facilitate filament stabilization by  In addition to stabilizing Rad51 filaments, Swi5-Sfr1 has been shown to stimulate the 5 ATPase activity of Rad51, which is also important for efficient strand exchange 20,21,23 . Since 6 substoichiometric concentrations of Swi5-Sfr1C failed to efficiently stimulate the ATPase 7 activity of Rad51 25 , we sought to determine whether Rad51-dependent ATP hydrolysis was 8 potentiated by the 7A mutant. As expected, wild type Swi5-Sfr1 was able to efficiently enhance 9 ATP hydrolysis by Rad51 at substoichiometric concentrations (Swi5-Sfr1:Rad51 ratio of 1:20), 10 with a 1.85-fold increase in ATP turnover (Fig. 6g). In contrast, the 7A mutant only managed 11 to enhance the ATPase activity of Rad51 1.28-fold, which is similar to the 1.27-fold stimulation 12 observed with Swi5-Sfr1C. The 3A and 4A mutants stimulated ATP hydrolysis like wild type. 13 These results suggest that interaction of either Site 1 or 2 with Rad51 is sufficient to promote 14 efficient stimulation of ATP hydrolysis. 15

DNA repair in Sfr1-Rad51 interaction mutants is facilitated by Rad51 paralogs 17
Collectively, the in vitro defects of the 7A mutant are highly reminiscent of observations made 18 with Swi5-Sfr1C (i.e., in the absence of Sfr1N). Since the sfr1C strain was as sensitive to DNA 19 damage as sfr1D ( Fig. 1d and Supplementary Fig. 1a,c), it seemed likely that strains 20 expressing Sfr1 mutant proteins defective in the interaction with Rad51 would also be sensitive 21 to DNA damage. To test this prediction, strains were constructed in which the native sfr1 + 22 gene was replaced with either sfr1-7A, sfr1-3A or sfr1-4A. Unexpectedly, the interaction 23 mutants did not show any obvious sensitivity to DNA damage ( Fig. 7a and Supplementary Fig.  24 5a). A marginal sensitivity was observed for the sfr1-7A strain but this was not statistically 25 significant (Fig. 7b). 26 To explain these results, we considered the possibility that, although the interaction 27 between Rad51 and Swi5-Sfr1-7A is compromised in vitro, Swi5-Sfr1-7A may still associate 28 with Rad51 in vivo via another Rad51-interacting protein. Previous genetic studies suggested 1 that there are two HR subpathways in S. pombe: one pathway is dependent on Swi5-Sfr1 and 2 the other pathway is dependent on the Rad51 paralogs Rad55-Rad57 17,19 . Although these two 3 pathways are thought to function in parallel, promoting Rad51 activity independently of each 4 other, it remained formally possible that Rad55-Rad57, which interacts with Rad51 11 , could 5 bind to Swi5-Sfr1-7A and serve as a molecular bridge to facilitate Rad51 stimulation. To 6 examine this possibility, the interaction mutants were introduced into the rad55D background. 7 Strikingly, in the absence of Rad55, the sfr1-7A mutant showed the same DNA damage 8 sensitivity as the sfr1D mutant (Fig. 7c). Furthermore, both the sfr1-3A and sfr1-4A mutants 9 were more sensitive to DNA damage than sfr1 + in the rad55D background, although this 10 sensitivity was not as severe as that observed for the sfr1D rad55D and sfr1-7A rad55D strains 11 ( Fig. 7c,d). Similar results were obtained in the rad57D background ( Supplementary Fig. 5b), 12 verifying the requirement for an intact Rad55-Rad57 auxiliary factor complex. Importantly, 13 these results suggest that defects in the direct interaction between Swi5-Sfr1 and Rad51 are 14 suppressed by a Rad55-Rad57-dependent mechanism that allows Swi5-Sfr1 to stimulate 15 Rad51 in vivo. 16

DISCUSSION 1
In this study, we characterized the interaction between Rad51, the key protein in HR, and 2 Swi5-Sfr1, a widely conserved recombination auxiliary factor. The N-terminal half of Sfr1 was 3 found to be essential for the role of Swi5-Sfr1 in promoting Rad51-dependent DNA repair (Fig.  4 1). This domain was shown to be intrinsically disordered (Fig. 2) and contain two sites that 5 interact with Rad51 (Fig. 3,4). Although mutation of the two interaction sites disrupted the 6 physical and functional interaction with Rad51 in vitro (Fig. 5,6), unexpectedly, defects in DNA 7 repair were only observed in the absence of Rad55-Rad57, another auxiliary factor complex 8 (Fig. 7). 9 10 Cooperative interactions with Rad51 are the primary function of Sfr1N 11 Although Sfr1N is not essential for stimulation of Rad51-driven DNA strand exchange (Fig.  12 1b,c) 25 , it was found to be essential for the promotion of Rad51-dependent DNA repair by 13 Swi5-Sfr1 ( Fig. 1d and Supplementary Fig. 1a-c). NMR interaction analysis revealed that two 14 domains within the N-terminus of Sfr1 interact with Rad51 ( Fig. 3 and Supplementary Fig. 3). 15 Mutation of Site 1 or 2 weakened the interaction with Rad51 while mutation of both sites 16 resulted in a near-complete loss of interaction ( Fig. 5b and Supplementary Fig. 4c), indicating 17 that Sites 1 and 2 of Sfr1N bind cooperatively to Rad51. Interestingly, Rad51-driven strand 18 exchange and ATP hydrolysis were significantly impaired only when both sites were mutated 19 (7A mutant, Fig. 5c-f and 6g). These results indicate that the reduced interaction in the single 20 site mutants (3A or 4A) is sufficient for Swi5-Sfr1 to fully stimulate Rad51 in these assays. 21 Although this points towards some functional redundancy, it is possible that these assays were 22 not sensitive enough to detect marginal defects in the stimulation of Rad51-driven strand 23 exchange and ATP hydrolysis. Indeed, the fluorescence anisotropy assay revealed a severe 24 defect in Rad51 filament stabilization for the 7A mutant and a modest defect for the 3A and 25 4A mutants (Fig. 6b-f), indicating that interaction of both Sites 1 and 2 with Rad51 is important 26 for efficient filament stabilization. Swi5-Sfr1 has also been shown to stabilize Rad51 filaments against the F-box helicase Fbh1 37 . It would be interesting to test whether the interaction 1 mutants can function in a similar capacity. 2 The reduction in NMR signals from Sites 1 and 2 in the presence of Rad51 (Fig. 3e), 3 combined with the gradual chemical shifts observed for some residues (Fig. 3c,d), indicated 4 that the association and dissociation of Sfr1N and Rad51 is fast on the NMR timescale, 5 suggesting that the Sfr1N-Rad51 interaction is relatively weak. Like all RecA-family 6 recombinases, S. pombe Rad51 is predicted to exist as a multimer in solution. Consistently, 7 size-exclusion chromatography yielded a broad elution profile with a retention time 8 corresponding to ~160 kDa ( Supplementary Fig. 6a). While the size of monomeric Rad51 9 (~40kD) is too small to cause severe line-broadening of NMR signals, the interaction of 10 multimeric Rad51 with Sfr1N could explain the drastic reduction in NMR signals for Sites 1 11 and 2. The NMR interaction analysis was largely substantiated by site-specific crosslinking of 12 residues within Sites 1 and 2 to Rad51 ( Fig. 4 and Supplementary Fig. 3e,f). Since both sites 13 are involved in electrostatic interactions with Rad51, the more robust crosslinking of Site 1 14 may be due to the added contribution of hydrophobic interactions between Site 1 and Rad51. 15 Alternatively, it is possible that the incorporation of the aromatic pBPA into Site 1 was better 16 tolerated due to the more hydrophobic nature of this site. 17 Since in vitro results obtained with the 7A mutant complex closely resemble what was 18 observed in the complete absence of Sfr1N (i.e., with Swi5-Sfr1C), we conclude that the 19 function of Sfr1N in promoting Rad51-dependent DNA repair primarily involves cooperative 20 binding of Sites 1 and 2 to Rad51. We nevertheless note that the 7A complex displayed a loss 21 of DNA binding and 4A showed impaired DNA binding ( Supplementary Fig. 6b). sfr1-3A is 22 more sensitive to DNA damage than sfr1 + in the rad55D background (Fig. 7c, indistinguishable in all other aspects (Fig. 5-7). Although we cannot completely rule out the 26 possibility that the defects of the 7A mutant are related to a defect in DNA binding, these 27 results strongly suggest that DNA binding by Swi5-Sfr1 is impertinent to its role in stimulating 28 Rad51 activity or promoting Rad51-dependent DNA repair. Mouse Swi5-Sfr1 (mSwi5-Sfr1), 1 which stimulates Rad51 through similar mechanisms (see below), does not display any DNA 2 binding 38 while human Sfr1 (hSfr1) has been implicated in transcriptional regulation 39 , raising 3 the possibility that the DNA binding activity of S. pombe Swi5-Sfr1 may have some relevance 4 to a function other than DNA repair. 5 6 Rad51 paralogs promote Swi5-Sfr1-dependent DNA repair 7 While the sfr1D and rad55D single mutants are sensitive to DNA damage, neither is as 8 sensitive as rad51D, which is epistatic to both 9,19 . However, because the sfr1D rad55D double 9 mutant shows the same sensitivity as rad51D (e.g., Fig. 7a) 17,19 , it was concluded that two 10 independent sub-pathways of HR exist in S. pombe 19 . Despite the numerous defects observed 11 in vitro, the sfr1-7A mutant strain was proficient for DNA repair, but this repair was completely 12 dependent on Rad55-Rad57 ( Fig. 7a-d), indicating that a Rad55-Rad57-dependent 13 mechanism overcomes defects in the binding of Swi5-Sfr1 to Rad51. 14 To explain these results, we propose that the interaction of Swi5-Sfr1 with Rad51 is 15 enabled by two redundant mechanisms: one through a direct interaction and the other through 16 Rad55-Rad57, which interacts with Rad51 11 and acts as a molecular bridge to facilitate the 17 recruitment of Swi5-Sfr1 to Rad51 (Fig. 7e). Hence, although Swi5-Sfr1-7A cannot interact 18 directly with Rad51, Rad55-Rad57 aids the recruitment of Swi5-Sfr1-7A to Rad51, allowing it 19 to exert a stimulatory effect on Rad51; this explains why sfr1-7A is proficient for DNA repair. 20 However, in the absence of Rad55 or Rad57, this tethering is lost but Swi5-Sfr1 can 21 nevertheless promote some DNA repair via its direct interaction with Rad51, thus explaining 22 why rad55D is not as sensitive as rad51D. It is only when both interaction mechanisms are 23 defective, as in the rad55D sfr1-7A strain, that the promotion of Rad51-mediated DNA repair 24 by Swi5-Sfr1 is completely lost. We therefore surmise that, while the Swi5-Sfr1 and Rad55-25 Rad57 sub-pathways are capable of operating independently of each other, as observed in 26 the rad55D and sfr1D backgrounds, Swi5-Sfr1 and Rad55-Rad57 likely collaborate to promote Rad51-dependent DNA repair in wild type cells. While physical evidence for this model is 1 lacking, Rad55-Rad57 facilitates recruitment of the Shu complex to Rad51 by binding to both 2 and acting as a molecular bridge 14,40 , so it could plausibly fulfill a similar role for Swi5-Sfr1. 3 However, unlike the Shu complex, Swi5-Sfr1 can interact directly with Rad51 ( Fig. 5b and  4 Supplementary Fig. 4c) 20,25 , so any contribution made by Rad55-Rad57 to this interaction 5 would enhance rather than enable complex formation with Rad51. The requirement for such 6 a mechanism may stem from the fact that the direct interaction between Swi5-Sfr1 and Rad51 7 is relatively weak (see above). Indeed, previous attempts by us and others to co-IP Swi5-Sfr1 8 and Rad51 from yeast extracts has been unsuccessful 19,41 , suggesting that the cellular 9 interaction is too weak and/or transient to capture. It is tempting to speculate that Rad55-10 Rad57, Swi5-Sfr1, and the Shu complex exist as a higher-order auxiliary factor complex, 11 perhaps as part of a Rad52-containing DNA repair center 42,43 . Evidence for the existence of 12 such a complex and elucidation of its molecular function will be the focus of future research. A consistent trend across all examined species is that Sfr1 plays some role in 8 facilitating the interaction with the recombinase partner. Since the amino acid sequence of the 9 N-terminal half of Sfr1 shows little conservation compared to the C-terminal half 10 (Supplementary Fig. 4a), it is tempting to ascribe the similarities among species to the C-11 terminus. However, sequence divergence across large evolutionary distances does not 12 necessarily reflect a lack of functional conservation for intrinsically disordered regions, which 13 accumulate mutations at a higher rate than structured domains 57 . Notably, the large subunit 14 of RPA contains an intrinsically disordered region whose function is conserved despite 15 significant sequence divergence 58 , raising the possibility that the structure and/or function of 16 Sfr1N is conserved. While empirical evidence is lacking, disorder predictions 59 for the N-17 terminal half of S. pombe Sfr1 agree with the data presented here ( Supplementary Fig. 7a) 18 and similar profiles were generated for Sfr1 from Schizosaccharomyces japonicus and 19 Schizosaccharomyces octosporus ( Supplementary Fig. 7b,c). Furthermore, the N-terminal 20 halves of mSfr1, hSfr1 and Mei5 are predicted to be enriched in intrinsically disordered regions 21 (Supplementary Fig. 7d-f). This analysis also highlighted potential protein binding sites within 22 the N-terminal halves of S. japonicus Sfr1, hSfr1 and Mei5. In agreement with this, the N-23 terminal half of Mei5 has already been shown to interact with Dmc1 52,55 . Thus, in addition to 24 the conserved function of Swi5-Sfr1 in promoting HR, the intrinsically disordered nature of 25 Sfr1's N-terminus and its role in facilitating interactions with recombinases may be 26 evolutionarily conserved. Further studies will be required to test the validity of this prediction.

ACKNOWLEDGEMENTS 1
We thank Tomohiro Koizumi for contributing to the early stages of this study; Yumiko 2 Kurokawa and Yuki Ide for help with protein purification; and Ryoji Miyazaki, Hiroyuki Mori and 3 Yoshinori Akiyama for help with the site-specific crosslinking experiments. This study was (yellow), and >80% (red) are highlighted, along with the corresponding residues. Underlined residues correspond to Site 1 (S84 to T114) and Site 2 (T152 to S168), where the most 1 significant signal attenuations were observed. Residues not in bold include prolines, 2 unassigned residues, and residues with overlapped signals.  For (b,d), mean values of three independent 12 experiments ± s.d. are shown. For (b), statistical analysis was by one-way ANOVA followed 13 by Tukey's multiple comparisons test. *, P < 0.05. n.s., not significant (P > 0.05). 14 ONLINE METHODS 1

S. pombe and E. coli strains 3
The genotypes of S. pombe strains used in this study are listed in Table S1. Standard media 4 was used for growth (YES), selection (YES with drugs or EMM), and sporulation (SPA), as 5 described previously 60 . The genotypes of E. coli strains used in this study are listed in Table  6 S2. Standard media was used for growth (LB) and selection (LB with antibiotics), unless 7 otherwise indicated. All strains employed in this study are available upon reasonable request 8 to the corresponding author (HI). 9 10 DNA damage sensitivity 11 A single colony was resuspended in 2 mL of YES and grown for 24 h (rad + ) or 48h (rad -). Cells 12 from these cultures were then seeded into 2 mL of fresh YES and grown for ~14 h until they 13 reached log phase (~0.8 x10 7 cells/mL). Cell density was adjusted to 2 x 10 7 cells/mL, 10-fold 14 serial dilutions were made, and 5 µL of each dilution was spotted onto YES plates (no 15 treatment control) or YES plates containing the indicated genotoxins. In the case of UV 16 irradiation, cells were spotted onto YES without any drug and treated with acute UV exposure 17 of the indicated dose. The leftmost spot on each plate contains 1 x 10 5 cells. Cells were 18 photographed with a digital camera after the indicated growth period (2-4 days). For 19 clonogenic assays, cells were grown as described above and spread onto several YES plates 20 and irradiated with the indicated dose of UV. After 3 (rad + ) or 4 (rad -) days of growth, colonies 21 were counted. 22

Extraction of cellular proteins for immunoblotting 24
Cells (1 x 10 8 ) were harvested and processed exactly as previously described 61 . Briefly, 25 harvested cells were resuspended in 1mL of ice-cold water. 150µL of lysing solution (1.85 M 26 NaOH 7.5% beta-mercaptoethanol) was added and mixed with the cells, followed by a 15 min 27 incubation on ice. 150µL of 55% TCA was added, followed by a further 10 min incubation on 28 ice. Precipitated proteins were pelleted by centrifugation (16,000 g 10 min 4˚C) and dissolved 1 with mixing (65˚C 10 min) in 100 µL of urea loading buffer (8 M urea, 5% SDS, 200 mM Tris-2 Cl pH 6.8, 1 mM EDTA, 0.01% BPB, freshly supplemented with 10% volume each of 1 M DTT 3 and 2 M Tris). Proteins were separated by SDS-PAGE, transferred to PVDF membranes, and 4 detected with the indicated antibodies. To verify the signal assignments, the following samples were prepared and their 1 H-3 15 N HSQC spectra were obtained at 500 MHz; amino-acid selectively 15 N-labeled (Ala, Arg, 4 Ile, Leu, Lys, or Phe) Sfr1N(1-176), Arg-selectively 15 N-labeled R97A variant of Sfr1N(1-176), 5 Lys-selectively 15 N-labeled K93A variant of Sfr1N(1-176), and uniformly 15 N-labeled Sfr1 (127-6 176). From the 160 expected main-chain amide NH signals, 157 were detected (98%) and 7 assigned to specific residues in Sfr1N. The remaining three signals from Q2, S3, and H51 8 could not be assigned due to line-broadening. NMR data were deposited in the Biological 9 Magnetic Resonance Bank (BMRB) repository with accession number 27885. 10 11 Secondary structure analysis 12 The secondary structural elements were analyzed by calculating deviations of the observed 13 13 Cα, 13 Cβ, and 13 CO chemical shifts from their residue-dependent random coil values 30,70 . 14 Residues were deemed to form random coils if they displayed secondary chemical shift values 15 within a limited range (between -0.7 and 0.7 for 13 Cα and 13 Cβ atoms, and between -0.5 and 16 0.5 for 13 CO atoms of non-proline residues, and -4 to 4 for all three 13 C atoms of proline 17 residues). The program TALOS+ 31 was also used to predict the secondary structural units 18 where 1 HN, 13 Cα, 13 Cβ, 13 CO, and 15 NH chemical shifts were used as input data. Predictions of 19 disorder and protein binding sites for Sfr1 homologs were generated by DISOPRED3 59 . 20

Purification of proteins for biochemical analyses 22
Previously published protocols were followed to purify Rad51 21 , RPA 20 , and Swi5-Sfr1 (wild 23 type 20 , Sfr1N 24 , and Swi5-Sfr1C 24 ). Swi5-Sfr1 mutants (3A, 4A, 7A) were purified by the same 24 method as wild type Swi5-Sfr1 except they were diluted to ~25 mM NaCl instead of ~100 mM 25 NaCl before being applied to the HiTrap Heparin column. All proteins were free of 26 contaminating nuclease and ATPase activity for the duration of the relevant assays. In all assays where comparisons were made between reactions with and without protein, the 1 equivalent volume of protein storage buffer was added instead of the protein. and plotted, with the standard deviation of these averaged values depicted by error bars. In 26 Fig. 7b, a one-way ANOVA followed by Tukey's multiple comparisons test was performed. *, P < 0.05, n.s., not significant (P > 0.05). Further statistical information for Fig. 7b  Three-strand exchange assay 5 Following staining with SYBR Gold (Thermo Fisher Scientific), gels were imaged using a 6 LAS4000 mini (GE Healthcare). Densitometric analysis was performed using Multi Gauge 7 software (version 3.2, Fujifilm) exactly as described 20,21 . Background signal above the lds, NC 8 and JM bands was subtracted from the corresponding values. The JM value was divided by 9 1.5 to compensate for the extra signal generated by these three-stranded DNA molecules. 10 The sum of the values was set to 100%, and the percentage of total DNA corresponding to 11 NC or JM was calculated. For the total yield, the percentage of DNA corresponding to NC and 12 JM was combined. The values from three independent experiments were averaged and 13 plotted, with the standard deviation of these averaged values depicted by error bars. Raw data 14 is available upon reasonable request from the corresponding author (HI) 15 16