RecF protein targeting to post-replication (daughter strand) gaps II: RecF interaction with replisomes

Abstract The bacterial RecF, RecO, and RecR proteins are an epistasis group involved in loading RecA protein into post-replication gaps. However, the targeting mechanism that brings these proteins to appropriate gaps is unclear. Here, we propose that targeting may involve a direct interaction between RecF and DnaN. In vivo, RecF is commonly found at the replication fork. Over-expression of RecF, but not RecO or a RecF ATPase mutant, is extremely toxic to cells. We provide evidence that the molecular basis of the toxicity lies in replisome destabilization. RecF over-expression leads to loss of genomic replisomes, increased recombination associated with post-replication gaps, increased plasmid loss, and SOS induction. Using three different methods, we document direct interactions of RecF with the DnaN β-clamp and DnaG primase that may underlie the replisome effects. In a single-molecule rolling-circle replication system in vitro, physiological levels of RecF protein trigger post-replication gap formation. We suggest that the RecF interactions, particularly with DnaN, reflect a functional link between post-replication gap creation and gap processing by RecA. RecF’s varied interactions may begin to explain how the RecFOR system is targeted to rare lesion-containing post-replication gaps, avoiding the potentially deleterious RecA loading onto thousands of other gaps created during replication.


INTRODUCTION
The accurate replication of genetic information is an essential process allowing cell proliferation and genome stability. The replisome is a multienzyme complex formed by more than ten components. In bacteria, DNA replication starts from the origin, proceeds bidirectionally and ends at the terminus site ( 1 , 2 ). During replication, cells experience endogenous or exogenous stresses causing DNA damage. Encounters of the replication machinery with unrepaired DNA damage can lead to replication stalling or collapse ( 3 , 4 ). In some cases, encounters with template lesions do not halt r eplisome progr ess. Instead, the r eplisome can bypass the lesion and be reprimed downstream, leading to the formation of a lesion-containing post-replication gap (5)(6)(7). Single-strand DNA post-replication gaps are formed fre-quentl y, perha ps se v eral times per replication cycle under normal growth conditions ( 8 , 9 ). Despite decades of work, the formation and repair of post-replication gaps remains one of the least understood processes in DNA metabolism. Once formed, the gap will usually contain a lesion and accurate r epair r equir es an undamaged complementary strand. Post replication gaps can be filled by three different mechanisms: (i) homologous recombination (10)(11)(12), (ii) RecAindependent template switching (13)(14)(15)(16) and (iii) translesion DNA synthesis ( 11 , 17 , 18 ). RecA-dependent homologous recombination predominates.
To repair a lesion-containing post-replication gap, the RecA protein must find that gap and distinguish it from other gaps that occur normally during replication and do not r equir e r epair. Resear ch over the last 5 decades has implicated the RecF, RecO and RecR proteins in two relevant functions: (a) targeting repair specifically to lesioncontaining post-replication gaps and (b) loading RecA onto single-stranded DN A (ssDN A) within those ga ps through the displacement of the single-stranded DNA binding protein SSB (19)(20)(21)(22). The grouping of r ecF , r ecO and r ecR genes into an epistasis group has been substantiated by a range of genetic observations (23)(24)(25)(26)(27)(28)(29)(30)(31)(32). Howe v er, a stab le comple x containing all three proteins has not been observed. For this reason, the term RecFOR will be used here onl y w hen referring to the system or pathway. The loading function is embedded in the R ecO and R ecR proteins, which form a complex that is necessary and sufficient for loading RecA protein onto SSB-coated ssDNA in vitro and in vivo ( 20 , 21 , 33-35 ). RecO but not RecF binds directly to the C-terminus of SSB and RecOR loads RecA protein at more or less random sites within a ss-DN A ga p ( 10 , 33 ). The targeting function we wish to explore in the current study appears to be centered on RecF.
RecF is a member of the ATP-binding cassette proteins (ABC) and harbors the Walker A, Walker B and the Signature domains characteristic to this superfamily ( 36 , 37 ). Structurally, RecF is similar to structural maintenance of chr omosome pr oteins (SMC) and notably to the head domain of Rad50, a eukaryotic ABC protein involved in double strand break repair ( 36 , 37 ). RecF forms a dimer in which two ATP molecules are located at the interface created between the Walker A and signature domains of two opposite monomers. In vitro , RecF protein binds single strand (ss) ( 38 ) and double strand (ds) DNA. The binding to DNA is highly ATP dependent ( 37 , 39-41 ). RecF exhibits a weak DNA-dependent ATPase acti vity gov erning its dissociation from the dsDNA ( 42 ). Ther efor e, RecF binding to dsDNA can be enhanced if ATP hydrolysis is blocked, either by using the non-hydrolyzable A TP analogue A TP ␥ S or a RecF mutant protein lacking ATPase function such as RecF K36R ( 42 ).
RecF also makes a complex with RecR ( 41 , 43-45 ). Both R ecO and R ecF compete for R ecR binding ( 45 ). A more stab le RecF-dsDNA comple x is formed when the RecF dimer is stabilized through interaction with RecR protein to form the RecFR complex, although in that case an incr ease in ATP ase acti vity is also observ ed re v ealing faster recycling ( 40 , 41 , 43 ). RecF or RecFR will also bind to ss-DNA ( 37 , 38 , 40 , 41 , 43 ).
The most prominent hypothesis proposed to date for RecF targeting to post-replication gaps envisions specific binding of RecF and / or RecFR to the ends of gaps ( 12 , 20 , 46-49 ). This hypothesis has become unwor kab le for at least two reasons. First, strong binding by RecF to gap ends would presumably be problematic, as there would be no way to distinguish the occasional lesion-containing postreplication ga ps w here RecF activity is needed and the much more plentiful gaps generated by lagging strand DNA synthesis, along with much less common gaps produced by mismatch repair and other processes. Unneeded and potentially deleterious DNA joint molecules linking the sister chromosomes could be generated behind the replication fork. Second and importantly, RecF exhibits no strong binding preference for a DNA end or a ds / ss junction at the end of a gap ( 40 , 42 , 50 ), a status established systematically by the accompanying manuscript ( 40 ).
The mechanism by which the RecFOR system is producti v ely targeted to lesion-containing post-replication gaps is thus unclear. If RecF is involved in targeting but does not bind specifically to gap ends, the targeting function must be found in other RecF interactions. An interaction with one or more proteins found at or near post-replication gaps becomes an important possibility. The most obvious interaction candidates are SSB and the replisome. RecF does not interact with SSB (10, 21, 33, this work).
In addition, the promiscuous RecA loading function of RecOR seen in vitro must somehow be constrained in the cell so that RecA filaments are not loaded into gaps that do not r equir e r epair. Constraining RecOR-mediated loading of RecA protein means blocking RecO interactions with RecR, SSB, or both. RecF can enhance the RecOR-mediated loading of RecA onto SSB-coated ss-DNA ( 46 , 47 , 50 ), but to date this has only been observed under two conditions in vitro where the interaction of RecO protein with ssDNA-bound SSB is blocked. As described further in the Discussion, the details of experiments that detected a RecF stimulation of RecOR function might reflect a RecR handoff between RecF and RecO.
Recent studies have begun to phenotypically distinguish the RecF and RecO proteins that might reflect the RecF (targeting) and R ecO (R ecA loading) division of labor within the R ecFOR system. R esistance to particular DNA dama ging a gents is more dependent on R ecF than R ecO and vice versa (51)(52)(53). After DNA damage or replication fork stress, only RecO and RecR are essential for RecA f oci f ormation in Bacillus subtilis ( 51 ). Similarl y, after DN A damage, R ecO and R ecR, but not R ecF, ar e r equir ed for nucleoid compaction observed in E. coli ( 53 ). In contrast, of the RecFOR proteins, only the RecF protein is toxic to cells when ov er-e xpr essed (54)(55)(56). This r epr esents a key and dramatic distinction between R ecO and R ecF. Moreover, RecF's deleterious effect is suppressed by RecOR coexpression, suggesting a compensatory effect ( 55 ).
Another factor distinguishing R ecF from R ecO is a growing literature, based on both experimentation and speculation, linking RecF protein to a function at the replisome ( 44 , 52 , 57-60 ). The recent ability to visualize singlemolecules in living cells has demonstrated that RecF and RecO do not colocalize and exhibit very different spatiotemporal behavior ( 52 ). Whereas RecO is generally found at sites distal to the replisome, RecF often colocalizes with the replication fork ( 52 ). RecF is required for rapid resumption of DNA synthesis after cells are UV irradiated and pre v ents e xtensi v e DNA degradation from occurring ( 58 , 61 , 62 ). These functions r equir e the RecF ATP ase ( 62 ). Intriguingly, the recF gene is located adjacent to dnaN in an operon otherwise devoted to replication, an evolutionary rela tionship tha t could be accidental but has ne v er been rationalized ( 63 , 64 ).
Could a RecF interaction with the replisome explain the targeting of the RecFOR system to post-replication gaps? Post-r eplication gaps ar e cr eated when a r eplisome encounters a lesion and disengages from the template. With a RecFreplisome interaction, proper placement of RecF for repair purposes could, in principle, be coupled to replisome disengagement to create the gap requiring repair. In this report, we explore the function that the RecF protein might have at the replication fork. We document an interaction between RecF and replisome components (particularly DnaN), investiga te how RecF af fects replisomes, and provide evidence suggesting that replisome destabilization is at the heart of the toxicity seen when RecF concentration is increased. When combined with studies already published, the observations may help explain how lesion-containing postreplication gaps are distinguished from other gaps, how RecF protein finds its way to the particular ga p w here it is needed, and how the potentially deleterious RecA-loading function of RecOR might be constrained.

Strains and plasmids
All the strains and plasmids used in this study are listed in the Tables 1 and 2 below. Strain were constructed using RED recombination ( 65 ) or P1 transduction as indicated. All constructions were confirmed by PCR and sequenced as r equir ed.
For the EAW1130 (P BAD -recF ) and EAW1148 (P BAD -recF K36R ) E. coli K-12 MG1655 deri vati v e strains, the araBAD promoter was inserted in the front of the start codon (ATG) of recF or recF K36R to replace the nati v e promoter. Briefly, a sequence containing a transcription termina tor a t the end dnaN followed by the araBAD promoter in the front of the recF or recF K36R genes was cloned into a plasmid and then amplified by PCR. For all constructs, PCR fragments were gel purified and integrated onto the chromosome using RED recombination ( 65 ). For recF-mKate2 mutants the promoter region of gyrB was duplicated to maintain the gyrB promoter region. This maintained normal gyrB expression (Figure 2 B).

Cloning
The lacIZ region of the pRC7 vector was amplified by PCR using the following primers, BsmI / lacIq up2: 5 -CGGA TAGAATGCGCAA TTCGGGACA CCATCGAATGGTGCAAAAC and BsmI/lacZ rev2: 5 -CGGATAGAAT GCGT GTTTTTTAAATAGTACA TAA TGGA TTTCCTTA. The PCR product was DpnI digested for 1 h, gel purified, and digested with BsmI in order to be ligated into a pBR322 plasmid linearized by BsmI and dephosphoryla ted. The liga tion product was transformed into DH5 ␣ competent cells ( lac -). After one hour of growth, the transformed cells wer e spr ead on plates supplemented with ampicillin. The resulting vector is pEAW1232.
The open reading frame (start to stop codons) of the following genes: recF , recF K36R , recR , recO , recG , dnaC , dnaE , dnaN , dnaG , topB and recQ were amplified and subcloned in frame into pGAD-C3 and pGBD-C3 to generate N-terminal fusion with GAL4 either the activation domain (AD) or binding domain (BD). The resulting plasmid were attributed a pEAW number (see plasmid list) and identified as pGAD-protein of interest or pGBD-protein of interest.
The RecF-mKate2 was subcloned from pEAW1128 into pET21A vector by double digestion EcoRI and NdeI. Finally, the plasmid encoding His-mKate2 was generated as followed. The pEAW1290 (recF-mKate2 in pET21A), was digested with BglII and EcoRI and the small DNA band containing the linker-mKate2 region, was ligated to pET28A cut with BamHI and EcoRI. The resulting plasmid was directly sequenced to confirm it was linker-mKate in pET28A.

Media and culture condition
Chemicals and media wer e pur chased from Fisher, Sigma, Biolabs, AlphaAesar or Teknova. Cells were grown in rich media Luria Bertani (LB), or in EZ supplemented with 0.2% glycerol. L -Arabinose (Ara) was purchased from Fisher (Acros Organics), 20% stock was made by resuspending the Ara in ultrapure water and filter sterilization. When r equir ed, antibiotics wer e added at the following concentrations: ampicillin 100 g / ml, kanamycin 40 g / ml and tetracycline 15 g / ml.

Over -expr ession assay
The RecF wt and its ATPase dead variant RecF K36R unlabelled or labelled ov ere xpression was carried out in different conditions to study new aspects of RecF function. For protein ov er-e xpr ession from plasmids, cell cultur es wer e inoculated with 1:100 ratio of an overnight (ON) culture grown in the same condition, i.e. in LB (or EZ glycerol) supplemented with ampicillin. Cells were grown at 37 • C to midlog phase OD 600 : 0.2-0.4, then 0.2% of Ara was added to induce ov er-e xpr ession. For the microscop y experiments, all experiments were carried out in EZ 0.2% glycerol. When the chromosomal constructions were used for ov er-e xpression, cultur e wer e inocula ted with a 1:1000 ra tio from ON cultur e, then cells wer e grown for 16 h at 37 • C with various concentr ations of Ar a (0 to 10%). In order to use high concentr ations of Ar a, a 2x LB was pr epar ed and mix ed with the adequate volume of 20% Ara and completed with ultrapure sterilized water to reach the final volume.
Survival was assayed after Ara addition at the indicated time by spotting assay. Briefly, cells were serial diluted in phospha te buf fered saline (PBS 1x, pH 7.4) by a factors of ten and 10 l of the indicated dilutions were spotted on LB plates (supplemented with ampicillin for cells carrying pBAD or pBR322 deri vati v e v ectors). Plates were incubated overnight a t 37 • C , then images were taken using a LAS (GE Healthcare) or iBright (ThermoFisher) imagers.

Protein detection by coomassie staining or immunoblot
The production le v el in total cell extracts of a variety of proteins of interest from this study was determined by Coomassie stained SDS PAGE gel and / or immunoblot. Cells were harvested at the indicated time. Cell pellets were dir ectly r esuspended in adequate volume of cracking buffer (CB) composed of 10% glycerol, 125 mM TrisCl pH 6.8, 2% SDS, 5% 2-betamercaptoethanol and 0.5 mg / ml Bromophenol blue. The volume of CB added was adjusted to the OD 600 (the volume use was calculated to be equivalent to 100 l of CB for a pellet of 1 ml of cells at OD 600 :1 Membranes were washed in PBS-T 4 times for 3 min, then incubated 1h in PBS-T with the second anti-body diluted 1:5000 (Goat anti Rabbit HRP coupled). After incubation, membranes were washed four times 3 min in 1xPBS in order to be re v ealed using the ECL kit (SuperSignal West Pico Plus, Thermofisher) into an iBright imager. Western Blot from independent membranes as cells extract needed to be more concentrated in order to properly detect GyrB were used.
Representati v e biological r eplicates ar e pr esented.

Gr owth curv e and SOS induction
Growth curves and SOS induction assays were carried out on cells either deficient in or ov ere xpressing RecF variants to evaluate the impact on growth and increase in DNA damage of the different conditions. Strains were transformed with plasmids pQBI63 (for empty vector control) or pEAW903 ( PrecN-sfgfp ) . Cell cultures were started with an inoculum 1:1000 of a sa tura ted culture into 100 l of LB amp containing the indicated arabinose concentration and the mix was transferred into a costar black microplate with 96 wells. The microplate was loaded into a Synergy microplate reader (Biotek) set at 37 • C with shaking. The OD 600 and the sfGFP fluorescence signal was measured e v ery 10 min for 16 h. Finally, the fluorescence signal was calculated for each condition at each time point as followed: ( GFP cells pEAW903 ) / A600 cells pEAW903 − ( sfGFP cells pQBI63 ) / A600 cells pQBI63 In the case of the growth restart assay, we only followed the OD 600 . Strains were cultivated as previously described with increased concentration of arabinose for the first 16h, then a dilution 1 to 1000 was used to inoculate a fresh culture in LB amp onl y. Onl y the OD 600 is reported in this case.
Values indicated on graph are the mean and standard error of biological triplicates.

Imaging of live and dead cells
We used LIVE / DEAD BacLight (Molecular Probe) to assay the viability of cells following RecF ov ere xpression. After 16h of culture in the presence of the indicated concentration of arabinose, cells were spun down, washed and resuspended in 0.85% NaCl in order to be incubated with the adequate solution allowing the differential staining of li v e and dead cells as described by the manufacturer (Molecular Probe). Following the incubation, imaging of cells was For easiest dif ferentia tion of li v e and dead cells on the images, the LUTs of DsRed and GFP channels were respec-ti v ely changed to yellow and blue before merging the three channels.

Single-cell fluorescence microscopy imaging
Single-cell fluorescence imaging was used to study the behavior of RecF variants upon ov ere xpression alongside labelled replication proteins of interest: DnaX-YPet and SSB-mTur2 in living cells. All imaging experiments were realized in EZ defined medium (Teknova) supplemented with 0.2% glycerol and ampicillin (EZ glycerol amp) to minimize the auto-fluorescence observed with LB media. Cultur es wer e inoculated from overnight cultur e with 1:100 ratio and grown at 37 • C to reach mid-log phase as described earlier ( 52 ).
To carry out the experiments, cells were loaded into a home-built flow cell as described previously ( 52 ). Cells were briefly settled (2-5 min) to allow them to stick to a silanized coverslip (treated with 3-aminopropyl triethoxysilane APTES), then fresh medium was flowed in at the rate of 50 l / min using a syringe pump (Adelab Scientific) to both dislodge loosely associated cells and provide nutrients. During the experiment time course, freshly oxygenated medium was continuously flowed into the chamber incuba ted a t 37 • C . Cell positions were randomly determined in the bright field during the cell settling time. Time zero of the experiment corresponds to the first image capture. Directly after the capture of the first frame of each position, the flow was briefly stopped, and the EZ glycerol amp medium was substituted for the EZ glycerol amp supplemented with 0.2% arabinose. For all fluorescence imaging, an initial brightfield image of the cells was recorded (34ms exposure).
Rapid acquisitions (movies of 300 × 50 ms frames, continuous excitation with 568 nm light and collected between 610-680 nm with an ET 645 / 75m filter, Chroma) were realiz ed to characteriz e the motions of RecF-mKate2 and RecF K36R mKate2 molecules before arabinose addition. Two-color time lapse movies were recorded to visualize mKate2 fusion along with the replisome marker (DnaX-YPet) or the fluorescent fusion of single strand binding protein (SSB-mTur2) during RecF ov er-e xpression. These movies were used to determine the number of foci and the colocaliza tion pa ttern of the two fluorophores used in each of those e xperiments. RecF-mK ate2 was imaged using yellow excitation light ( λ = 568 nm) at high intensity (76.6 W cm −2 ) and collected between 610-680 nm (ET 645 / 75m filter, Chroma, at EM gain = 150, 100 ms exposure). DnaX-YP et w as imaged using green excitation ( λ = 514 nm) at higher laser power (287.1 Wcm −2 ) and collected between 525-555 nm (ET540, 30m filter, Chroma, at EM gain = 150, 200 ms exposure). Lastly, SSB-mTur2 was imaged using green excitation ( λ = 458 nm) at lower laser power (15.64 W cm −2 ) and collected between 468 and 495 nm (ET 485 / 30m filter, Chromaat EM gain = 200, 100 ms exposure). For all experiments, even when the untagged RecF was used, images were recorded in the mKate2 channel.

Image processing
Image analysis was performed in Fiji / ImageJ, using the single-molecule biophysics plug-in ( 78 ), the Grid / Collection stitching plug-in ( 79 ), custom macros, and MATLAB. In Fiji, raw ND2 images were converted to TIF format, prior to background correct and image flattening as previously described ( 52 ). MicrobeTracker 0.937 MAT-LAB plug-in was used to create cell outlines as regions of interest (ROIs). Manual outlines were used to ensure that only non-overlapping, in-focus cells were selected for analysis. Cell metrics such as cell length, area, and volume were also generated utilizing this plug-in. Cell outline ROIs were then imported into Fiji to aid in the extraction of additional cell metrics including mean cell intensity, cell lengths, and foci per cell. Note that cell outlines that are occasionall y imported improperl y fr om Micr obeTracker to ImageJ ( < 10 for each experiment) were excluded. For all time lapse e xperiments, indi vidual analysis of each replicate was realized separately, then data from separate analyses were combined. For rapid acquisition, the analysis of two sets of equivalent number of frames from a biological triplicate were analyzed separately.
Colocalization analysis of RecF with SSB-Tur2 and DnaX-mKate2 was determined using a previously described colocalization protocol ( 80 ). Briefly, focus centroid positions were obtained using Peak Fitter plug-in in Fiji / Ima geJ (discoidal avera ging filter of 1-4 if not mentioned or 1-3 for DnaX-YPet), then corrected for drift between fluorescence channels. Only foci with centroid positions located within 2 pixels (218 nm) of each other were classified as colocalized. Colocalization frequencies were then estimated as the ratio of colocalized foci to the total number of foci present at each time point.
Fiji tools were used to generate ROIs around RecF-mKate2 and SSB-mTur2 features under ov er-e xpression conditions. A discoidal averaging filter was first applied to stitched fluorescence channel stacks to reduce signal associated unbound protein / cellular auto-fluorescence. A binary mask was then generated using the Yen Thresholding algorithm with a set threshold matching that used with Peak fitter. The Analyze Particles tool was then used to generate ROIs around areas of positi v e intensity with areas ≥3 pixel 2 . ROIs were then applied to the original flattened stitched image stack to extract feature parameters such as area and mean intensity.

Tet recombination assay
Tet recombination assays were used to study the effect of RecF ov ere xpression and deletion on RecA-independent template switching e v ents. Cells transformed with plasmids carrying 101 bp Tet repea ts separa ted by various interspace lengths (pSTL74, pSTL78 or pMB302) were grown for 16 h in LB Amp media supplemented or not by the indicated concentration of arabinose. Cultures were serially diluted in PBS by factors of ten and the appropriate dilutions were spread on LB plates supplemented with Tet and / or Amp. After 16 h incubation at 37 • C, colonies were counted to determine the number of Tet e v ents (Tet / Amp) or the c.f.u (Amp). The percentage of Tet e v ents was determined by the frequency of e v ents relati v e to the c.f.u. and expressed in percentage. A minimum of 6 biological replicates were utlilized for each strain. The significance of the difference observed was tested by t-test for each sample relati v e to the wild type for the same condition.

Plasmid DNA electrophoresis
The state of plasmid DNA (intact supercoiled or smearing) purified from cells ov ere xpressing RecF or not was tested by electrophoresis. Two stains were utilized to determine whether in some conditions, the signal is increased using a DNA stained with stronger affinity to ssDNA compared with ethidium bromide. Cells carrying pBR322 or pSTL78 were grown in LB Amp supplemented or not with 10% arabinose for 16 h. Three to six mL of cells were harvested and resuspended in 600 l of water. Plasmids DNA were extracted using the PureYield Plasmid Miniprep System from Promega. Purified plasmid DNA of each sample was resuspended in ultrapure nuclease free water. The DNA concentration was determined by the absorbance at 260 nm using a Nanodrop. For each sample, 250 ng of DNA was resuspended in 1x GED Buffer (glycerol, EDTA and bromophenol blue) and loaded onto a 0.8% TAE agarose gel. After electrophoresis, and staining with ethidium bromide or SYBR Gold, gels were imaged using a Typhoon-FLA imager (GE Healthcare).

Electr on micr oscopy
The proportion of ds-versus ssDNA of plasmid DNA purified from cells ov ere xpressing RecF or not was determined by electr on micr oscopy. Samples for electr on microscop y (EM) wer e obtained by spr eading the r eaction mixtures with the cytochrome technique described previously ( 81 ). The technique allows the dif ferentia tion of ss versus ds DNA region based on the thickness of the DNA molecules observed. The plasmid DNA samples were first purified by minipreparation extraction followed by a phenol-chloroform extraction and ethanol precipitation to ensure the high purity of the sample. Samples were dialyzed against 20 mM NaCl and 4 mM EDTA for at least 16 h at 25 • C on Millipore type VM filters (0.05 m). Then samples wer e spr ead as described ( 81 ). Imaging and photography were carried out with a TECNAI G2 12 Twin Electron Microscope (FEI Co.) equipped with a 4k × 4k Gatan Ultr ascan CCD camer a. Digital images of the DNA molecules were taken at ×30 000 Magnification. DNA molecules were manually counted and sorted into groups as indicated in the figures.

Plasmid loss assay
The effect of RecF ov ere xpression on plasmid replication in living cells was tested by plasmid loss assay. Cells deleted of the lac operon (EA W408, EA W1400 and EAW1401) were transformed with the pEAW1232 or pRC7 plasmids and spread on LB plates supplemented with amp, 0.5 mM IPTG, 80 g / ml X-Gal in order to select cells carrying the plasmids. Transformed cells were then grown overnight in presence of the appropriate antibiotics before starting the experiment. Cell cultur es wer e started in LB supplemented or not by 10% arabinose with 1:1000 ratio of the saturated culture. The number of cells carrying the pRC7 or pEAW1232 plasmids were determined at time zero and after 16 h of culture in the absence of antibiotics. Cells were serially diluted in PBS by factors of ten. The adequate dilutions wer e spr ead on XGal IPTG plates and incubated overnight a t 37 • C . Finally, the blue and white colonies were counted to determine the plasmid loss for each str ain. Photogr aphs of the blue / white colonies plates were kindly taken by Robin Davies from the MediaLab of the Biochemistry department of UW Madison.

Yeast two hybrid assay
Interaction between RecF and partners was tested by Yeasttwo hybrid. First, yeast CFy7 cells were transformed as described earlier ( 82 ) with appropriate plasmids to test the interaction between RecF fused to one domain (activator AD or binding BD) and the indicated partner fused to the other encoded to the complementary plasmid pGAD or pGBD. Yeast transformants were selected on Leu-/ Trp-selecti v e drop out medium plates at 30 • C. Then 4 to 5 yeast transformants were grown overnight at 30 • C in liquid selective drop out medium. The optical density of overnight yeast culture was measured, and 1 mL of cells was harvested. Yeast cells were broken down by 3 cycles of freeze / thaw consisting of 3 min in liquid nitrogen and 3 min at 42 • C. Cells pellets wer e r esuspended in 1 mL of Z buffer (Na 2 HPO 4 60 mM, NaH 2 PO 4 40 mM, KCl 10 mM and MgSO 4 1mM) and ␤galactosidase assay was carried out as described ( 83 , 84 ). Biological replicates of 4 or 5 experiments wer e r ealized and significati v e difference relati v e to the negati v e control were tested by Mann-Whitney.

Protein pr epar ations
The E. coli RecF, RecF K36R and RecFmKate2 were purified as previously described ( 42 ). The E. coli RecR was purified using a dual-tag purification system allowing the purification of a protein of interest by a first step of maltose binding protein affinity purification, followed by the cleavage by the SUMO protease Ulp1 between the MalE-6His-Smt3 and the protein ( 85 ). This left the cleaved RecR protein without any tag as described earlier ( 40 ). For use as a control, the His-mKate2 was purified from 6L of LB amp culture of BL21 DE3 slyD transformed with pEAW1300 (6His-mKate2) induced at OD 600 : 0.4 with 0.5 mM IPTG and ov ergre w at 33˙C for 3h. Dry cell pellet of ∼20 g was flash frozen until cell resuspension. The cell pellet was resuspended overnight at 4 • C in lysis buffer (50 mM Tris-Cl pH 7.5, 100 mM NaCl, 20 mM imidazole, 10% glycerol). Lysis buffer was adjusted to 75 ml. Lyzozyme 1.25 mg / ml final concentration resuspended in fresh lysis buffer was added to the cell resuspension and the mixture was stirred gently for ∼1 h. Cell resuspension was sonicated 10 times with a cycle consisting of 1min sonication, set with 0.5 s on, 0.5 s off. The cell lysate was centrifuged for 60 min at 12 000 rpm, 4˙C using JLA. 16.250 Beckman Coulter rotor. Cell soup supernatant was loaded on gravity column of ∼10 mL Nickel r esin pr e-equilibra ted in Lysis buf fer. Column was washed with 3 column volume of lysis buffer prior to elution with 3 × 5 mL of elution buffer (same as lysis buffer but 500 mM Imidazole). Elution fractions were pooled and dialyzed against B buffer (50 mM Tris-Cl pH 7.5, 50 mM NaCl, 0.1 mM EDTA, 10% glycerol, 1 mM DTT). Protein was loaded on a 5 ml SP HP column and eluted on linear gradient to buffer D (same as B but 1M NaCl). Pooled fractions were dialyzed back into B buffer and flowthrough a 5 mL Q FF. Purified protein was dialyzed against RecF stora ge b uf fer and stored a t −80 • C .
The E. coli SSB protein was purified as described earlier ( 74 ). The E. coli replication enzymes: the replicati v e DNA polymerase PolIII, the clamp loader, the ␤-clamp unlabelled Nucleic Acids Research, 2023, Vol. 51, No. 11 5721 and labelled, DnaB, the RNA primase DnaG and MBP-AF647, were generous gifts from Dr S Jergic, Dr S Chang, Dr Richard Spinks and Dr N Dixon. All protein preparations were tested and found free of endo-or exonuclease activities. Protein concentrations were determined by absorbance at 280 nm using extinction coefficient of the monomeric form of each protein (if not specified otherwise),

Far western dot blot assay
Far western blot was used to test interaction between purified RecF and other purified proteins. Interaction between RecF and identified partners was confirmed by Far-western blot, using an adapted protocol described by Walsh et al. ( 86 ). Three microliters of two-fold serial dilution in RecR stora ge b uffer of protein partners RecR, BSA, DnaN and DnaG were spotted in duplicate (one use for the dot blot the other as control) to get 54 to 1.7 pmol of each protein on nitrocellulose membranes and dried at room temperature for 15 min. Membranes were blocked with 5% milk in PBS-T for 45 min at room temperature. Blocking solution was discarded, one membrane was incubated with 5% milk PBS-T containing 0.2 M of purified RecF overnight at 4 • C while the other was incubated with same volume of fresh blocking solution only (No RecF). Membranes were washed 4 times 3 min with PBS-T in order to be incubated for at least 3h with the anti-RecF antibodies (1:1000) in blocking solution. Membranes were washed 4 times 3 min with PBS-T and then incubated with the Goat-anti-Rabbit HRP antibodies in PBS-T for at least 2h. Membranes were washed 4 times 3 min with PBS before being re v ealed sim ultanousl y using ECL F anto. F ar w estern blots w ere carried out in biological quadruplicate and the quantification was obtained by subtracting the background signal observed in the control membrane to the signal obtained in the far-western blot membrane.

In vitro single-molecule interaction assay of labelled protein
This method was previously described and used to confirm the lack of exchange between labelled replicati v e proteins post PolIII complex formation ( 87 ). We used this method to validate the interaction of labelled proteins in a mixture. Purified labelled proteins wer e mix ed (40nM of mKate2 deri vati v es with 80 nM of AlexaFluor647 labeled) in 1x replicati v e buffer as described below (in the description of the rolling circle assay) and incuba ted a t 37 • C for 20 min. The mixed sample was then diluted 500 to 1000x in 1x replicati v e buffer and 50 l was dir ectly spr ead on a slide cleaned with 5M KOH and imaged immediately following the spreading. Imaging was realized with a Nikon Ti2-E (100x Objecti v e) equipped with EM-CCD camera (C9100-13, Hamamatsu) and a heated stage insert as previously described ( 77 ). Excitation was provided using semi-diode lasers of wavelengths 568 nm (Sapphire LP, Coherent, 27 mW max. output) and 647 nm (OBIS, Coherent, 38 mW max. output). Continuous imaging of 50ms images was first carried out in the 647 channel f or 1min, f ollowed by a continuous imaging of 50ms images in the 568 channel. In vitro single-molecule interaction experiments were carried out at three times for RecFmKate2 and DnaNAF647 and at least twice for the control. The analysis was carried using the ImageJ / Fiji softwares. Fields of views were first flattened, then average projection of the first 150 images (647 channel) or 50 images (568 channel) were generated. Individual foci in each channel were picked using the Peak fitter with constraints of 4 pixels fit radius, a minimum distance of 3 pixels between peaks and a discoidal averaging 1 3 was applied. Tables of peaks (foci) were corrected for the offset between channels and the corrected table were used to analyze the colocalization of the foci in both ways using a maximum distance between centroid of 3 pixels. Picked foci were then ordered as not colocalized if the distance was > 218nm or colocalized is the distance the distance between centroid was ≤ 218 nm. The coincidental chance of colocalization (C) between the two-colour foci was calculated as using the formula: where A R = area of the focus, n = number of foci (of both 647 and 568 channels), A FOV = area of the field of view. The distance between colocalized foci was used to sorted them by colocalization shell area, as described earlier ( 80 ).

Conserv ation structur e and sequence analysis
Sequence conservation of Esc heric hia coli RecF was generated using the AlphaFold structure of RecF (AF P0A7H0 F1) with Consurf server, conservation analysis was set to 300 sequences.

Multimers prediction structure
AlphaFold (see text) was used to predict the structure of potential multimers. AlphaFold was set up to predict the fiv e best models of the potential multimers RecF:DnaN (1:1, 1:2 and 2:2).

Single-molecule rolling-circle assay
Single-molecule rolling circle assay was used to study in real time in vitro replication in presence of the RecF variants. The dye used in the assay allows labelling of the newly synthetized double strand DNA and leave the single strand DNA unlabeled. Flow cells were prepared as described previousl y ( 88 ). Briefly, r eplication r eactions wer e carried out in microfluidic flow cells constructed from a PDMS flow chamber placed on top of a PEG-biotin-functionalized coverslip. Once, assembled with inlet and outlet tubing the flow cell was blocked against all non-specific binding by introducing blocking buffer (50 mM Tris-HCl pH 7.6, 50 mM KCl, 2% (V / V) Tween-20).
In vitro single-molecule microscopy was carried out on an Eclipse Ti-E inverted microscope (Nikon, Japan) with a CFI Apo TIRF 100 × oil-immersion TIRF objecti v e (NA 1.49, Nikon, Japan). The temperature was maintained at 33 • C by an electronically heated flow-cell chamber coupled to an objecti v e heating jacket (Okolab, USA). NIS-elements was used to operate the microscope and the focus was locked through Perfect Focus System (Nikon, Japan). Images were captured using an Evolve 512 Delta EMCCD camera (Photometics, USA) with an effecti v e pixel size of 0.16 m. DNA molecules stained with SYTOX Orange were imaged with a CW 568-nm Sapphire LP laser (200 mW max. output), and ET600 / 50 emission filter (Chroma, USA) at 0.76 W / cm 2 .
Reactions were carried out in replication buffer containing 25 mM Tris-HCl, pH 7.6, 10 mM magnesium acetate, 50 mM potassium glutamate, 40 g / ml BSA, 0.1 mM EDTA, 5 mM dithiothreitol, and 0.0025% (V / V) Tween-20. Conditions for the pr e-assembly r eplication r eactions wer e adapted from published methods (89)(90)(91). Solution 1 was pr epar ed as 30 nM DnaB 6 (DnaC) 6 , 1.5 nM biotinylated circular 2 kb dsDNA substrate and 1 mM ATP in replication buffer. This was incubated at 37 • C for 3 min. Solution 2 contained 50 M dCTP and dGTP, 6 nM 3 ␦␦' , 20 nM Pol III core ( ␣ε ) and 40 nM ␤ 2 in replica tion buf fer (without dATP and dTTP). Solution 2 was added to an equal volume of solution 1 and incubated for 5 min at 37 • C. This was then loaded onto the flow cell at 100 l / min for 1 min and then 10 l / min for 10 min. The flow cell was washed with replica tion buf fer containing 60 M dCTP and dGTP. Replication was initiated by flowing in the replication buffer with addition of 1 mM ATP, 250 M NTPs, 50 M dNTPs, 40 nM ␤ 2 , 75 nM DnaG, 100 nM SSB 4 , and 150 nM SYTOX Orange. W here indica ted 20 nM RecR, 10 or 100 nM RecF and 10 nM RecF K36R were used. All in vitro single-molecule experiments were carried out at least three times. Image analysis was performed in FIJI, using the Single Molecule Biophysics plugins (available at https: //github.com/SingleMolecule/smb-plugins ).

Primer extension assays
Primer extension assay were used as second method to study the impact of RecF addition on in vitr o replica tion. Primer e xtension e xperiments were carried out as described earlier ( 92 ), with the following modifications. Reactions were carried out in 40 mM Tris-HCl pH 7.2, 20 mM magnesium chloride buffer in which fresh dithiothreitol was added at the final concentration of 10 mM. When mentioned, RecF (or RecF K36R ) and RecR proteins wer e r especti v ely added last at 40 and 80 nM before starting the reaction. Otherwise, r eactions wer e carried out by mixing 1 mM ATP, 500 M dNTPs, 30 nM clamp loader ( 3 ␦␦' ), 90nMPol III cores ( ␣ε ), 200 nM ␤ 2 , and 750 nM SSB 4 . For the leading lagging replication reactions 75 nM DnaG, 250 M NTPs were also added. Reaction mixtures were kept on ice before starting the reaction. The addition of 6 ng of primed DNA was used to start the replication reactions, which were then incuba ted a t 30 • C . Aliquots of 10 l were harvested a t 0, 5 and 40 min. The reaction was stopped by the addition of the equal volume of the SDS, EDTA loading buffer prewarmed at 42 • C. Finally, samples were loaded onto 0.66% Agarose TAE gel, submitted to electrophoresis and SYBR Gold stained.

Softw ar e
ImageJ / Fiji (Microscopy) and Adobe Photoshop (Plates and gel) were used to edit the images. Brightness and contrast were uniformly adjusted for all images compared. Cells were manually outlined, to select single cells in focus with the MicrobeTracker tool in MATLAB 2013. MicrobeJ was used to automatically outline and classify Li v e and Dead cells ( 93 ). Excel, Origin, PRISM and MATLAB software were used to edits and analyzed the data. Figur es wer e pr epared in Adobe Illustrator.

RESULTS
RecF was recently identified in a screen for proteins that become highly toxic upon ov er-e xpression due to an increase in DNA damage ( 56 ). This effect does not extend to overexpression of RecO or RecR. In the case of RecF, the toxicity had been noted previously and depends on its ability to hydrolyze ATP ( 54 ). The observed toxicity of RecF when the protein is present at higher-than-normal le v els is the jumping off point for the current study. Howe v er, we also further examine the effects of a recF deletion, the effects of physiological concentrations of RecF on replisome action in vitro , and the interaction of RecF protein with replisome proteins DnaN and DnaG. The experiments to follow were aimed at further defining RecF effects on replication forks and more broadly to explain the phenotypic distinctions between RecF and RecO.

Toxicity of RecF over-expression constructs
The effects of RecF ov er-e xpr ession wer e studied at the single cell le v el using unta gged and ta gged versions of RecF. Normal functionality of a RecF-mKate2 fusion encoded a t the r ecF chromosomal locus was demonstra ted previously ( 52 ). Here, a pBAD vector system was used to upregulate production of RecF wild type and mutant proteins plus tagged versions of all of these ( Figure 1 ). To validate our ov er-e xpression tools, complementation and the toxicity of the different versions of R ecF (R ecF, R ecF K36R , R ecF-mKate2 and RecF K36R -mKate2) were tested under growth conditions adapted for single cell imaging (EZ medium containing glycerol as carbon source). Briefly, a recF deletion mutant strain was transformed with vectors encoding the different variants of RecF (Figure 1 A). The functionality of the RecF-mKate2 construct was again validated by a UV sensitivity complementation assay, under conditions in which no arabinose was added for induction but in which leaky expression provided low levels of RecF protein (Figure 1 B). The RecF-mKate2 construct was able to complement the recF null mutant at the same le v el as a similar construct expressing wild type RecF. As expected, plasmids expressing the untagged or tagged version of ATPase-dead RecF K36R did not complement.
The le v els of toxicity induced by ov er-e xpression of the various pBAD constructs were then analyzed after addition of 0.2% ar abinose (ar a) to the culture. RecF toxicity is observed 30 min after induction using an agar plate-based spot assay. The toxicity is similar for the untagged and tagged versions of the wild-type protein with a 4-log decline in The functionality of RecF-mKate2 was determined by complementation assay supported by the leaky expression of recF permitted in absence of arabinose. Cell cultures were serially diluted by a factor 10 down to 10 −6 and serial dilutions were spotted in replicate on a LB amp pla te. Replica tes were then exposed to increased UV doses as indicated. Values on the top of the plates r epr esent the order of the dilution. C RecF ov er-e xpression toxicity assay. RecF ov er-e xpression was initiated by the addition of 0.2% arabinose. Cells were serially diluted and spotted on LB amp plates at time 0, 30 and 60 min after induction. survival (Figure 1 C). As observed in earlier studies, the untagged RecF ATPase (K36R) mutant produced no toxicity. A partial toxicity is observed after 1h of induction for RecF K36R -mKate2. To ensure that the ATPase dependency is not a consequence of a difference in expression level, the e xpression was e xamined both by Coomassie gel and Western blot (Supplementary Figure S1). We noticed that the e xpression le v els of the tagged versions are slightly lower than the untagged versions. In both cases the expression of the ATPase dead mutant (RecF K36R ) is similar (or slightly higher) compared with the wtRecF. Altogether, these results confirm that the toxicity of the tagged RecF is comparable to that of the untagged protein upon ov er-e xpression. The results also confirm that the toxicity relies on the RecF ATPase activity and is not a nonspecific effect of the overexpression of this particular protein.
To further investigate the effect of RecF ov er-e xpression, we constructed a series of strains in which ov er-e xpression was mediated from the chromosomal recF locus. The nati v e recF gene is located in a complex operonic structure composed of dnaA-dnaN-r ecF-gyrB regula ted by multiple promoters distributed throughout the operon ( 63 , 64 ) ( Figure  2 ). The positioning of recF within an operon dominated by genes expressing proteins involved in replication has always been a curiosity. Notably, recR is also found in an operonic structure dominated by dnaX-ybaB (genes encoding respecti v ely two subunits of the clamp loader and a small DNA binding protein) ( 94 ). Of course, operon placement is not determinati v e. The g yrB gene is predominantly expressed as a single gene utilizing a promoter located in the a transcription termination sequence followed by the promoter of the ara BAD operon (P BAD ) was inserted at the locus in front of the start codon (ATG) of the gene encoding r ecF (or r ecF K36R ). After the r ecF stop, a Kan cassette followed by a duplication of the 3 end of recF carrying the gyrB promoter region was introduced (Figure 2 A). This arrangement preserved normal expression of the gyrB gene (Figure  2 B), unaffected by subsequent arabinose additions. A strain with the nati v e recF gene, in its normal operon context, was used as control in all the experiments carried out with these ov er-e xpression constructs. Reasoning that the chromosomal construct would produce lower RecF protein le v els, we carried out arabinose titration to determine the concentration exhibiting a toxicity similar to the plasmid construct in Figure 1 (Figure 2 and Supplementary Figure S2). To determine the toxicity of RecF ov er-e xpression, w e follow ed population growth with both OD 600 and colony forming unit (c.f.u.) measurements. Whereas almost no change in OD was observed, the c.f.u. decreased after arabinose addition. In the absence of arabinose, no toxicity was detected on plates ( Figure 2 and Supplementary Figure S2). About 1 and 1.5 log loss of survival was observed at 0.5 and 1% ara but the toxicity drastically increased to ∼2 and 3.5 log loss of survival when 5 or 10% ara were added, respecti v ely. The amount of RecF leading to toxicity (1 log loss of survival) corresponds to an increase of RecF ∼14 × (Supplementary Figure S2) which is estimated to be ∼700 molecules per cell compared to the initial le v el of 50 molecules per cell. No toxicity was observed for RecF K36R at the same concentrations of arabinose.
Western blots demonstrated that the production le v el of RecF was similar (or somewhat lower) to that of RecF K36R under these ov er-e xpression conditions (Supplementary Figur e S2). Mor eov er, western b lot anti-RecF carried out a t dif ferent times suggested tha t the maximum production of RecF le v el is reached at 6h with the higher dose of 10% ara (Supplementary Figure S2). We further used western-Blot anti-RecF to estimate the number of RecF per cell after 16h of culture (Supplementary Figure S2). To determine whether the difference observed between absorbance and c.f.u. was due to the effect of filamentation on absorbance readings or to the inability of cells to resume growth after RecF ov er-e xpression, we performed li v e and dead single-cell imaging and followed the growth restart of cells w hich previousl y ov er-e xpressed RecF ( Supplementary  Figures 3 and 4). Li v e and dead single cell imaging re v ealed an increase in cell length and cell death upon RecF overexpr ession (Supplementary Figur e S3). The maximum cell death detected is about 30% after addition of 10% arabinose, which is expected to be higher based on the 0.01% survival (c.f.u.) of a culture which experienced almost no decrease in absorbance. Howe v er, the growth restart assay (Supplementary Figure S4) re v ealed a delay of about 4 h for cells which previously e xperienced RecF ov er-e xpression. Overall, the results of RecF over-expression from the chromosome replicate the previous observations of RecF overexpression toxicity from a plasmid and further suggest that RecF ov er-e xpression toxicity is due in large measure to a flaw in growth restart when RecF is ov er-e xpressed.
Two sets of published studies differ in the levels of SOS induction observed as a result of RecF over-expression ( 54-56 ). Sandler et al. ( 54 , 55 ) observed a decrease in SOS induction following UV or mitomycin C exposure, monitoring a short 2 h window following RecF ov er-e xpression (using a temperature inducible plasmid system). Conversely, Xia et al . ( 56 ) detected an increase in SOS induction after 24h of RecF ov er-e xpression (using an IPTG inducib le plasmid system). In the present study, strains containing chromosomally expressed RecF or RecF variant transformed with a P recN-sfgfp fusion expressed on plasmid were used to assay SOS induction. We first analysed the SOS le v el upon RecF ov er-e xpression, both alone and combined with UV stress to address the apparent difference in SOS induction previously observed (Supplementary Figure S5). In the first 2h following the ov er-e xpression a mild delay in SOS induction was observed for the RecF over-expression strains. Consistent with Sandler's findings, the le v el of SOS induction was relati v ely low at early times ( 54 , 55 ). Later, the SOS response became prominent (Supplementary Figure S5), as seen by Xia et al. ( 56 ). Ther efor e, we propose that the difference between previous studies is likel y primaril y due to the timing of the SOS experiments and possibly also to a difference in the RecF ov er-e xpression induction system utilized.
We further analysed the effect of RecF ov er-e xpression using increased arabinose concentration (in the absence of UV) by monitoring both the fluorescence deri v ed from the SOS-induced GFP and overall cell growth. The mean fluorescence observed for cells with the native recF promoter varies from 0 to a maximum of 3000 A.U. after 16h with arabinose (Figure 2 D). In the absence of arabinose, the RecF ov er-e xpression construct with the P BAD promoter, exhibited similar results. However, under increased overexpression conditions of 10% arabinose, the SOS-mediated GFP expression strongly increased > 20 times after 9 h. For the ATPase dead RecF K36R inducible construct, in the absence of inducer, a modest SOS signal is observed after 16h. This small induction of the ATPase dead mutant mimics the SOS le v el observ ed with the same fusion upon loss of the recF gene in absence of exogenous stress ( 52 ). In the presence of arabinose, this modest le v el of SOS induction decreased to background le v els seen in experiments with the cells carrying the nati v e recF promoter.
Finally, we tested the SOS induction in the first 16 h of RecF ov er-e xpression with 10% arabinose in homologous recombination deficient mutant strains, recA , recB , r ecO or r ecR (Supplementary Figure S6). Upon RecF ov er-e xpression, no change in SOS induction was observed in the recR strain. Significant delays in SOS induction were observed for r ecO and r ecB while no induction was observed in the r ecA nega tive control. Altogether, these results confirm the toxicity of RecF ATPase ov er-e xpression and re v eals its correlation with both SOS induction and a defecti v e cell capacity to return to growth. The results suggest an increase in DNA damage and generation of ssDNA that is dependent on RecF ATPase.

RecF o ver -expr ession incr eases r eplisome dissociation
Reasoning that RecF acts in some manner near the replisome, we investigated the effect of RecF ATPase ov er-e xpr ession on r eplisome stability in vivo using a fluorescence-based imaging approach. We set up a two-color imaging strain carrying a replisome marker (DnaX-YPet) along with RecF-mKate2 expressed from the araBAD promoter with addition of 0.2% arabinose ( Figure  3 ). DnaX-YPet and RecF-mKate2 signals were respecti v ely characterized at 514 and 568 nm by imaging individual cells and foci therein. As high le v els of mK ate2 could be excited at 514 nm and ther efor e cr eates a possible channel overlap of signal, we first determined if any artifactual mK ate2 signal (b leed through) could be observ ed on the YPet channel (Supplementary Figure S7). Control imaging was carried out with one-color strains expressing only RecF-mKate2. Images were recorded in both the YPet and mKate2 channels under over-expression conditions. At time 0, no artifactual signal was detected in the YPet channel. Howe v er, an artifactual signal in the YPet channel appeared after 60 min of RecF-mKate2 over-expression. Based on this result, the imaging time-lapse of DnaX-YPet in the two-color strains was limited to the first 30 min after induction of R ecF-mKate2 / R ecF K36R -mKate2.
No significant increase of mKate2 cellular fluorescence was observed in two-color strains expressing either RecF-mKate2 or RecF K36R -mKate2 30 min after arabinose induction. Nonetheless, pre vious western b lots re v ealed increased RecF-mKate2 production during that time-period. Though at first glance this observation appears contradictory, we attributed these observational differences to limitations of the mKate2 fluorophore. The fluorophore mKate2 was previously determined to have a half ma tura tion time of roughly 20 min ( 95 ). Thus, during our observational window of 30 min, it is reasonable to assume that the maturation lag of mKate2 fluorophores could obscure the observation of newly created RecF-mKate2 protein. Ther efor e, most of the RecFmKate2 foci observed likely result from the basal leaky expression and limit us to track only a part of the RecFmKate2 pool. Importantly, RecFmKate2 overexpr ession r esults r e v ealed a similar toxicity relati v e to RecF alone, suggesting tha t ma tura tion of the mKa te2 doesn't affect the RecF portion of the fusion. Next, we determined the number of mKate2 and replisome foci per cell (Figure 3 C and Supplementary Figure S8). Before induction, strains expressing either RecF-mKate2 or RecF K36R -mKate2 exhibited similar numbers of replisome foci per cell ∼1.7 (top panel). Howe v er, the number of mKate2 foci (RecF) was significantly smaller for the ATPase dead RecF K36R mutant with 0.06 ± 0.02 (RecF K36R -mKate2) versus 0.66 ± 0.06 (RecF-mKate2) (bottom panel). This suggests that the ATPase function may be needed for RecF dimerization and DNA binding in vivo . A similar result was obtained when acquiring rapid-acquisition movies rather than time-lapse series (Supplementary Figure S8). After arabinose addition, the number of RecF-mKate2 foci slightly decreases after 10 min of induction whereas it increases to 0.21 ± 0.02 for RecF K36R -mKate2 (Figure 3 C bottom panel and Supplementary Figure S9). The use of a replisome marker (DnaX-YPet) re v ealed that RecF ov er-e xpr ession corr elates with a sharp decline in replisome foci, beginning at 10 min after induction and decreasing further from 20 to 30 min ( Figure  3 C top panel). Over 70% of the visible replisome foci disappear upon ov er-e xpression of the RecF-mKate2. A much more modest decline is observed upon ov er-e xpression of the ATPase deficient RecF K36R -mKate2.
The proximity of the RecF-mKate2 and RecF K36R -mKate2 foci to the replisome was further analyzed by examining histograms of pairwise-colocalization distances. To account for the fact that short distances are sampled less frequently in these types of radial-search measurements, the histograms of colocalized foci were binned by area shells, as opposed to linear distances ( 52 ). The number of mKa te2 / YPet colocaliza tion counts in close proximity was higher for RecF-mKate2 at time 0 and remained high for the first 10 min (Supplementary Figure S9). During the first 10 min after arabinose addition, 14.43 ± 1.16% of replisome foci colocalized with RecF-mKate2) (Figure 3 D). This colocalization declined within 30 min coinciding with a decline in the total number of replisome foci. The colocalization is more substantial if considered from the standpoint of the less common RecF foci. For both RecF variants, a significant le v el ( ∼40%) of the RecF foci colocalized with replisome foci (Figure 3 D), although the numbers of RecF K36R -mKate2 foci wer e low. Ther e wer e fe w mK ate2 f oci to f ollow, v ery fe w replisome f oci included them < 2% f or the first 10 min and this number further declined within 30 min. Throughout the 30 min of the experiment, RecF K36R -mKate2 foci remained relati v ely rare. Howe v er, we noticed tha t when RecF K36R -mKa te2 f ormed f oci, the proximity to the replisome was not different from that seen for RecF-mKate2. Thus, RecF f ocus f ormation exhibits a strong dependence on the RecF ATPase activity, whereas the proximity of RecF foci to the replisome does not.
To confirm some of the key observations over a longer period of time, we next imaged the single-color dnaX-YPet strains, ov er-e xpressing the dark (untagged) versions of R ecF and R ecF K36R (Figur e 4 ). Overall patterns r emained the same. The relati v e total YPet concentration per cell was similar for both strains during the 60 min observation window, suggesting similar concentrations of replisome proteins (at least DnaX). RecF ov er-e xpression again produced a significant decrease ( > 70%) in replisome foci ( Figure 4 and Supplementary Figure S10). The number of replisome foci observed at 30 min was similar to that seen in the previous imaging of the two-color strains, for untagged RecF and RecF K36R respecti v ely. The decline continued from 30 to 60 min post-induction of the experiment for both constructs, reaching 0.43 ± 0.03 replisome foci for RecF and 0.97 ± 0.04 for R ecF K36R . R ecF protein ov er-e xpression thus leads to replisome uncoupling and transient destabilization in a reaction that is largely dependent on an intact RecF ATPase activity.
In principle, replisome dissociation could hav e se v eral dif ferent ef fects on the local binding of SSB: (i) a reduction caused by RecA protein loading onto the ssDNA region mediated by RecOR on the abandoned fork, with coincident SSB removal ( 96 , 97 ); (ii) a static presence of SSB if the replication is resumed by the r eplication r estart proteins without further DNA unwinding or (iii) an increase in the ssDNA SSB coated region, if the abandoned replication fork is further processed by helicases or if post-replication gaps are formed. To explore these possibilities and follow the fate of SSB, we used a new SSB-mTur2 visualization tool de v eloped by Keck and cowor kers ( Figure 5 ) ( 67 ) to image the ssDNA regions (i.e. replisome and gap). Unlike other SSB fusions studied to date, E. coli cells grow  normall y w hen SSB-mTur2 is the onl y SSB expressed. Controlling for possible channel overlap with mKate2 (RecF) (Supplementary Figure S7), we detected no artifactual foci in the mTur2 channel (458 nm). Strains expressing chromosomal SSB-mTur2 alongside of RecF-mKate2 or RecF K36R -mKate2 were imaged for 60 min after arabinose addition ( Figure 5 ). In agreement with the number of replisome foci observed under the same growth conditions, the number of SSB-mTur2 foci before induction was around 2 foci per cell for both strains. In contrast to the replisome, the number of SSB foci increased slightly after induction. As might be expected, this suggests that a region of ssDNA remained whether the replisome was present or not. When the ATPase dead RecF K36R -mKate2 was expressed, a similar small increase in SSB foci was observed to reach ∼2.5 after 60 min of induction. Prior to induction, low le v els of RecF-mKate2 foci (about 0.5 per cell) and very few RecF K36R -mKate2 foci wer e pr esent. Both RecF-mKate2 and RecF K36R -mKate2 foci increased upon induction, mainly after 30 min as the newly synthesized mKate2 fluorophor es matur ed. RecF-mKate2 strongly colocalized to the SSB foci. Although RecF K36R -mK ate2 e xhibited many fewer foci after 30 min, these also colocalized well to the SSB f oci. Even bef ore induction, 71.4 ± 3.3% of RecF-mKate2 foci colocalized with SSB-mTur2 foci. The colocalization slightly increased at 30 min, and then returned to the initial colocalization le v el. Colocalization of the detectab le RecF K36R -mK ate2 foci with SSB foci was significant (33.3 ± 12.6%) but lower than with RecF-mKate2 protein. In contrast to the replisome colocalization analysis, this implies that RecF might bind more often near nonreplisomal SSB foci than RecF K36R . Upon induction, as RecF K36R -mKate2 increased and formed more visible foci after the 30 min mark, its colocalization with SSB-mTur2 foci reached a plateau between 50 and 60%. In the reciprocal analysis, 14.3 ± 1.4% of the SSB-mTur2 foci contained RecF-mKate2 foci prior to induction and less than 1% of the SSB-mTur2 foci colocalized with the much smaller number of RecF K36R -mKate2 foci. After induction, the fraction of SSB foci colocalizing with either RecF variant increased substantially with ma tura tion of mKa te2 fluorophores after the 30 min mar k. Ov er 89% of the SSB foci colocalized with RecF-mKate2 after 60 min and just under 50% in the strains expressing RecF K36R -mKate2, perhaps partially due to the presence of fewer RecF K36R -mKate2 foci.
Finall y, the anal ysis of the SSB foci characteristics analyzed as particles in Fiji re v ealed that RecF ov er-e xpression increased the size and brightness of the particles in a . SSB-mTur2 was more often found colocalized with RecF-mKate2 than RecF K36R -mKa te2. The colocaliza tion incr eases after 30 min with a gr ea ter ef fect for RecF-mKa te2. ( C , D ) Analysis of the SSB particles (particle = continuous region of individual or overlapping SSB foci). The area (C), and the intensity (D), of more than 700 particles were determined at time 0, 30 and 60 min and are represented as a scatter plot. manner that was much more pronounced for strains expressing the RecF-mKate2 protein (Figure 5 CD). The size and brightness of RecF foci (particles) also increased after 30 min, although this may simply reflect the slow maturation of the mKate2 fluorophore (Supplementary Figure  S11). Altogether, these data show that RecF ov er-e xpression does not greatly affect colocalization with SSB foci but it does affect SSB feature size and brightness as well as RecF DNA binding. These results again indicate an increase in ss-DNA generated by RecF ov er-e xpr ession, corr elating with a decline in cellular replisome numbers.

RecF o ver -expr ession stimulates r epeat deletion events associated with post-replication gaps
The colocalization behavior of RecF associates the protein with both replisomes and gaps. If RecF ov er-e xpression is leading to larger numbers of post-replication gaps, then it might also lead to an increase in recombination e v ents linked to those gaps. This experiment utilized an assay developed by Lovett and collaborators ( 15 , 71 , 72 ), measuring deletion e v ents between nearby short (101 bp) dir ect r epeats on plasmids that are largely RecA-independent and strongly associated with post-replication gaps ( 71 , 72 ). The deletion e v ents create tetracy cline r esistance and ar e r eadily selected for. We carried out experiments with three plasmids harbouring variously sized regions between the two tet repeats of 101bp homology (1.4, 7.1 kb and cruciform formed by a palindrome of 110 bp) (Figure 6 A). Recombination e v ents betw een repeats w ere detected by plating after 16h of culture following induction by arabinose addition. No protein tags wer e pr esent on the R ecF or R ecF K36R proteins. For all of the assayed plasmids, significant increases in deletion e v ents were seen when the wild type RecF protein was induced. Incr eases wer e minimal or absent when the RecF K36R protein was induced or when the wild type recF promoter (unresponsi v e to arabinose) was used.
To expand the correlation and examine conditions that did not involve RecF over-expression, the assay was then repeated in strains lacking RecF protein (Figure 6 B). In agreement with the observations of Lovett and co-workers on intermolecular recombination e v ents between tet repeats greater than 50 bp ( 14 ), a recF deletion in all cases decreased the frequency of the e v ents. The same result was obtained for strains expressing the ATPase dead RecF K36R . When using a plasmid in which the r epeats ar e separated by 7.1 kb (where the background rate of deletion is very low), a deletion of the recO gene also caused a measurable decline in deletion frequency. When a plasmid was used that had much less DNA (110 bp with a long palindrome) separating the repeats, the loss of RecF function again caused a decline in deletion frequency (Figure 6 B). For this latter deletion substrate, where deletion frequencies are much higher, deletion of recO did not produce a decline in the manner of recF . In an attempt to confirm the recF and recO patterns, a similar set of experiments was carried out in a background in which the functions of the Uup and RadD proteins are missing. Eliminating these two genes has the effect of enhancing the recombination signal, as deletion of those two genes produces a synergistic increase in these types of RecA-independent dele-tion e v ents ( 8 ). The pa tterns seen with r ecF and r ecO were confirmed with these strains. As most of this recombination is RecA-independent, the effect of the recF deletion indica tes tha t RecF is involved, a t least in part, in a process that does not involve RecA protein loading into the gap. The more modest effects of the recO deletion are consistent with the role of RecO in RecA loading. In general, these experiments indica te tha t deletion e v ents associated with post-replication gaps increase when RecF is ov er-e xpressed in an ATPase-dependent fashion and decline when RecF is not present. The RecF ov er-e xpression appears to be associated with an increase in gap formation and / or an increase in gap size that provides fertile ground for RecA-independent recombination.

RecF o ver -expr ession incr eases damage and ssDNA f ormation on plasmid DNA
We reasoned that an effect on replisome stability, along with an increase in gap formation, might be especially detrimental to small replicons (plasmids) and might be reflected in an increase in DNA damage and plasmid loss. We specifically examined the effect of RecF ov er-e xpression on the stability of the plasmid pBR322 ( 70 ). We first took strains expressing RecF from its wild type promoter on the chromosome, as well as RecF and RecF K36R expressed from the pBAD promoter. Before and after addition of 10% arabinose to induce RecF or RecF K36R expression for 16h, plasmid DNA was purified, visualized and analyzed. The toxicity of RecF ov er-e xpression for the strains carrying the plasmid was controlled using an agar plate-based spot assay (Supplementary Figure S12). After plasmid purification by standard mini-preparation carried out in the same manner for all str ains, the DNA concentr ation was determined by the absorbance at 260 nm. In each case, 250 ng of DNA was loaded on two identical agarose gels (Figure 7 ). Following electrophoresis, identical gels were stained either by ethidium bromide or SYBR Gold. SYBR Gold is a more sensiti v e stain ab le to detect lower concentration DNA and single strand DNA. For most samples, the majority of the 250 ng of pBR322 DN A a ppeared to be supercoiled with both stains (Figure 7 A). The exception was the DNA purified from cells ov er-e xpressing RecF. In that case, the majority of the DNA is spread into a smear. Unlike the other samples, a larger amount of DNA was detected by SYBR Gold staining, thus suggesting a potential increase in ss-DNA. A similar smearing was observed for a larger plasmid (pSTL78; 7.1 kb) (Supplementary Figure S12). Much of the spontaneous DNA damage in cells is oxidati v e ( 98 ), an effect that is amplified when cells are grown in rich media with aeration (as is the case for most of the cell growth experiments in this study). We thus isolated plasmid pSTL78 from cells grown anaerobically. Ov er-e xpression of RecF again uniquely eliminated the duplex DNA circles from the plasmid prep (Supplementary Figure S12). All of these experiments were carried out at least 3 times with identical results. Together, these results suggest that RecF ov er-e xpression results in considerable damage to plasmid replicons with a potential increase in ssDN A. Unfortunatel y, the smear observed does not allow for dif ferentia tion between ss-and dsDNA formation. Tet deletions e v ents was tested upon increased concentration of arabinose. Wild type (P wt -recF ), EAW1130 (P BAD -recF ) and EAW1148 (P BAD -recF K36R ) strains were transformed with the pSTL78 (upper panel), pSTL74 (medium panel) or pMB302 (lower panel). ( B ). The r equir ement for RecF ATPase for the Tet repeats recombination e v ents was tested for deletion and point mutation strains using a plasmid-based assay. The wt, EAW629 ( recF ), EAW1190 (recF K36R ) , EAW114 ( recO ), ZJR04 ( radD uup ), EAW1063 ( radD uup recF ), CJH0115 ( radD uup recF K36R ), and EAW1064 ( radD uup recO ), strains were transformed with the pSTL78 (upper panel) or pMB302 (lower panel). Significant difference compared to the parental strain (wt) was tested by Mann-Whitney and are indicated in black (* for P = 0.05, ** for P = 0.005 or *** for P = 0.0005), an additional Kruskal-Wallis test was realized to compare the effect of increased concentration of arabinose for each strains and significance is indicated in grey. Purified plasmid DNA (250 ng) was loaded onto two identical 1% agarose gels. After electrophoresis DNA was EtBr or SYBR Gold stained. The upper panel r epr esent the gel images of r epr esentati v e e xperiments and lower panel r epr esent the aver age of the r aw intensity signal detected for the major band for a biological triplicate ±SD. The P values of significant differences between staining methods are indicated in grey and differ ences r elati v e to the wt strain in the same conditions are indicated respecti v ely in orange and pink for EtBr and SYBR Gold. ( B ) Electr on micr oscopy images of pBR322 purified DNA using the cytochrome C spreading method. The cytochrome C spreading allowed the dif ferentia tion between the dsDNA (black arrow) and the ssDN A (w hite arrow). RecF ov er-e xpression led to an apparent increase in observed ssDNA, concomitant to an almost complete loss of the circular dsDNA in the plasmid preps.
In an attempt to bypass this limitation, the DNA present in the purified samples (pBR322) was further analyzed by electr on micr oscopy using the cytochr ome C method ( 81 ). This method allows for the dif ferentia tion between ss-and dsDNA regions of a DNA molecule that can be quantified (Figure 7 B and Supplementary Figure S13). For all strains, in the absence of arabinose, the majority of the DNA observed was circular double stranded with a few molecules exhibiting small open regions of ssDNA (Figure 7 B and Supplementary Figure S13). Little change was observed when arabinose was added to the control strain with a non-inducible wt promoter (native promoter). Howe v er, when RecF was ov er-e xpr essed, the cir cular dsDNA essentiall y disa ppeared in the purified sample. The DNA molecules observed were largely either linear branched double stranded DNA with single-stranded regions or singlestranded DNA with small dsDNA patches. The DNA purified after RecF K36R ov er-e xpression was prominently double stranded circles but with a slight increase in long linear molecules and a minority of single-stranded DNA with short dsDNA patches. The EM analysis is consistent with the idea that RecF ov er-e xpr ession r esults in an increase in plasmid damage that either precludes plasmid isolation by the standard preparation or increases ssDNA in the plasmids.
Reasoning that gap formation occurring on plasmid DNA can e v entually lead to plasmid loss e v en with a relati v ely stab le multi-copy plasmid like pBR322, we e xamined the effect of RecF ov er-e xpression on plasmid loss using a much different and independent assay ( Supplementary Figures 12 and 14). A pBR322 vector in which lacIZ has been cloned (pEAW1232) was used for a blue / white color screen assay. To determine the number of cells losing the plasmid (indicated by colonies that are white), the plasmid was tr ansformed into str ains deleted of the lac operon so that all the lac genes are encoded by the plasmid (Supplementary Figure S12 C and Supplementary Figure S14). A modest but consistent increase in white colonies lacking plasmid was observed in cells exposed to RecF over-expression for 16 h compared to the control or the ATPase deficient RecF strain. About 10% of the cells entirely lost the multicopy pBR322 deri vati v e plasmid, but only upon ov er-e xpression of RecF. A similar effect was observed when the assay was carried out with the pRC7 plasmid (also encoding the gene for ␤-galactosidase), which has a lower copy number and is m uch more easil y lost ( 73 ) (Supplementary Figure S12D). The increase in plasmid loss confirmed that some kind of DNA damage that is deleterious to small replicons occurs upon RecF ov er-e xpr ession in an ATP ase-dependent manner. Together, the visualization and the analysis of the plasmid DNA, along with the plasmid loss assay, implicates RecF ov er-e xpr ession with r eplisome instability and an increase in ssDNA gap formation.

RecF interacts with the DnaN ␤-clamp and the DnaG primase
The data presented so far suggested a direct link between RecF and the replisome. We therefore sought more direct evidence for such an interaction. The interaction between RecF and putati v e replisome partners was investigated by the yeast-two hybrid assay in vivo (Figure 8 ). The interactions were probed by measuring ␤-galactosidase activity . The interactions between RecF and the two newly identified partners DnaN and DnaG were further corroborated by far-western blot (Figure 8 and Supplementary Figure  S15), a method used previously to identify DnaN interaction partners ( 86 ). RecR and BSA were used as positi v e and negati v e controls respecti v ely. RecR, BSA, DnaN, DnaG and SSB purified proteins were serial diluted and spotted 54 to 1.7 pmol on the membrane. After blocking, the RecF protein was added to a fresh blotting solution at the concentration of 0.2 M, before further incubation with primary and secondary antibodies. This allowed detection of pr otein-pr otein interactions with RecF used as prey. As expected, after visualization no signal was detected for BSA and SSB (Supplementary Figure S15) while signal was detected for RecR. Signals of the same order of magnitude as RecR were detected for DnaN and DnaG.
The interaction of RecF and DnaN was further validated by single molecule microscopy. This method was previously used to demonstrate the absence of polymerase core exchange in solution when pre-assembled with the clamp loader ( 87 ). Purified RecF-mKate2 and DnaN-AF647 were mixed. As a control, RecF-mKate2 was mixed with MBP-AF647 and DnaN-AF647 was mixed with His-mKate2 ( Figure 8 ) to make sure the results did not involve an anomalous interaction with the protein fusion components. Mixtur es wer e incubated for 20 min before imaging. For the controls, colocalization was around 10%. In contrast, RecFmKate2 and DnaN AF647 were found to be highly colocalized ( ∼60%). The analysis of the colocalization shell area is indicati v e indicati v e of a close interaction (Supplementary Figure S15). Together, these three pr otein-pr otein interaction assays confirm the interaction between RecF and DnaN and also indicate a potential interaction with DnaG.
The amino sequence analysis of RecF (Supplementary Figure S16) generated with Consurf ( 99-103 ) did not re v eal a clear clamp binding motif 'CBM' or alternati v e motifs previously shown in partners to directly interact with DnaN (104)(105)(106)(107)(108)(109)(110)(111)(112). Ne v ertheless, AlphaFold ( 113 , 114 ) was successfully used to predict potential heteromultimers, including a heterodimer (Supplementary Figure S16). All of the predicted structures positioned RecF interacting with DnaN with at least one contact involving the region surrounding the loop consisting of residues 167-171 of RecF. Howe v er, they do not all map RecF interacting with the cleft of DnaN.

In vitro , RecF triggers ssDNA gap formation during replication
E. coli replication can be reconstituted and characterized in vitro using purified replisome proteins and coupled to a primed rolling-circle templa te, a t both the ensemble and single-molecule le v el . The replication process can be monitor ed r especti v ely either in bulk using an electrophoresis gel or in real time using single molecule fluorescence microscopy ( 92 , 115 , 116 ). To determine the effect of RecF and its ATPase activity on replisome stability and function, we set up replication assays in which purified RecF protein was added at both a physiolo gicall y relevant concentration (10 nM) ( 52 , 117 ) and also at a higher concentration to mimic RecF ov er-e xpression (100 nM). In some experiments, RecR was also added at a 2:1 ratio relati v e to RecF (Figure 9 and Supplementary Figure S17), with concentration as mentioned in the caption.
The single-molecule experiment is presented here first. The experimental design of this assay involves the replication of a rolling-circle DNA substrate tethered to the surface of a flow cell. The newly synthetized double-stranded DNA is stretched by a continuously applied laminar flow and visualized in real time using SYTOX Orange, a stain specific to double-stranded DNA (Figure 9 A). Therefore, if single-stranded gaps are formed in the product strand, staining is discontinuous. In the experiments presented below, the replisome was pre-assembled onto the DNA template in solution. During replication a noticeable and concentra tion-dependent dif fer ence was observed in the fr equency of gap formation when RecF was added to the r eaction (Figur e 9 B and C). The basal fr equency of visible gaps formed (in the absence of RecF) is on average 0.015 ± 0.002 gaps per m DNA synthesized. This number increased modestly to an average of ∼0.027 gaps per m DNA synthesized when either RecR or RecF K36R proteins were added alone, respecti v ely. Howe v er, this number was increased to more than 0.042 ± 0.003 gaps per m DN A synthesized w hen 10 nM RecF was added. Gap formation increased to more than 0.059 ± 0.004 gaps per m DN A synthesized w hen both RecF and RecR were added together (Supplementary Figure S17). These results suggest that RecR and RecF K36R play a role in replisome impairment and uncoupling but the ATPase activity of RecF appears to be an important factor. Unlike the experiments using RecF ov er-e xpression in vivo , these e xperiments utilized mainly RecF concentrations consistent with normal in vivo RecF concentra tions estima ted to be in the range of 5-20 nM, equivalent to 18-68 RecF molecules per cell ( 52 , 117 ). When RecF was added alone at the higher concentration to mimic ov er-e xpr ession, the fr equency substantially increased to 0.191 ± 0.008 gaps per m DNA synthesized. This observation correlates well with the increase in replisome loss and uncoupling observed in vivo upon RecF overexpression. The analysis of the gap size also re v ealed that when R ecR or R ecF K36R were added alone, the average size of the few gaps formed incr eased. Inter estingly, the addition of wild type RecF has the opposite effect; gaps formed are smaller and a significant further decrease is observed for the combination of RecF and RecR. These observations suggest that the RecF ATPase activity might not only be involved in the gap frequency but also in the initiation of the downstream Okazaki fragment synthesis.
The gaps observed above must be occurring primarily on the lagging strand as the formation of leading-strand gaps would lead to termination of the rolling-circle replication reaction. To determine if the RecF ATPase effect on gap formation was lagging-strand specific, we set up ensemble primer-extension assays, on primed M13 DNA, allowing the replication of the leading strand alone or the leading and the lagging when the DnaG primase and the rNTPs were added ( Figure 9DE). In both cases, the addition of RecR or RecF K36R had no effect on replication; the intensity of the final product was similar to the intensity observed for the control (storage buffer). In contrast, a reproducible decrease in the final replication product was observed when 20 nM RecF was added. This effect was greater when RecF and RecR were added together. As the effect of RecF was seen in both replication assays, these results suggest an involvement of RecF ATPase in gap formation during the ongoing replication in both strands of the DNA, with perhaps a stronger effect on the lagging strand.

DISCUSSION
The mechanism by which the RecFOR system is targeted to lesion-containing post-replication gaps is not understood. The most prominent targeting mechanism proposed to date involves specific RecF binding to the ends of gaps. Howe v er, as detailed in the accompanying paper ( 40 ), neither RecF nor the RecFR complex have the needed specificity for binding to gap ends. If binding to the most notable structural feature of a ssDN A ga p --the ga p ends --cannot explain targeting, then a pr otein-pr otein interaction becomes the most likely alternati v e. Here, we demonstrate an interaction between RecF and replisome components that opens an experimental path to solving the targeting conundrum and more. In principle, by interacting directly with the replisome that first senses the template lesion, replisome disengagement to create a post-replication gap could leave RecF behind, properly positioned to direct repair of that particular gap. The overall scheme has some elements that mirror the quite detailed speculation put forward by Kuzminov 25 years ago ( 60 ).
The case we present for RecF interaction with replisomes and RecF effects on those replisomes has multiple and varied components. In brief, ( 1 ) the toxicity associated with RecF ov er-e xpr ession is her e associated with r eplisome destabilization; ( 2 ) there is a direct interaction of RecF with DnaN and possibly DnaG; and ( 3 ) RecF triggers gap forma tion in vitr o a t concentra tions found in vivo . All of this work complements a growing literature associating RecF with the replisome ( 44 , 52 , 57-60 , 62 ). We expand on these three conclusions below.
Ov er-e xpression of the RecF protein is highly toxic to a bacterial cell ( 54 , 56 ). To date, the molecular basis of that toxicity has not been understood. We conclude that overexpression of wtRecF, with its ATPase intact, directly and negati v ely impacts replisome stability, limiting the capacity of cells to resume normal growth following RecF overe xpression. RecF ov er-e xpression leads to dramatic cellular r eplisome loss, incr eased SSB particle number and size suggesting an increase in gap formation, increased recombination associated with post-replication gaps, and significant loss of small replicons (plasmids) e v en when they exist as Nucleic Acids Research, 2023, Vol. 51, No. 11 5737 multiple copies in the cell. The increase in gap-associated recombination, a large induction of the SOS response that is recO-and recB-dependent, and the loss of circular duplex plasmid circles all associate the RecF ov er-e xpression with replisome dissociation and an accompanying increase in ss-DNA that would be expected if gaps were being formed. These observations support our proposal that the destabilizing effects on replisomes underlie the toxicity associated with RecF ov er-e xpr ession. A dir ect interaction between RecF and the replisome provides a better explanation for the observed replisome destabilizing effects of RecF overexpr ession than r eplisome collisions with randomly bound R ecFR complexes. R ecF K36R , which binds to dsDNA as well or better than RecFR, has no toxic effects when overexpressed at the same or higher le v els. Unlik e RecF, o verexpression of RecO has no toxic effects. We thus propose that the effects of RecF ov er-e xpr ession r eflect a RecFreplisome interaction and resulting replisome destabilization that explains the toxicity associated with RecF overexpression. We of course acknowledge that over-expression of any protein can create anomalies and that the results must be considered in that context.
The interaction between RecF and the replisome is readily demonstrated. A previous screen using a global pulldown assay to study new pr otein-pr otein interactions identified a potential interaction between RecF and DnaN ( 118 ). Here, we document a direct interaction between RecF and DnaN with three distinct methods. Yeast two hybrid experiments and far western blots documented the interaction between RecF and DnaN, along with a possible but much weaker interaction with DnaG. A strong co-localization of RecF and DnaN can also be observed by microscopic examination of single molecules, providing a third result confirming the interaction. These observations strongly suggest a direct interaction between RecF and the replisome that can explain the co-localization observed in vivo under normal growth conditions ( 52 ).
Finally, we demonstrate that addition of RecF protein to acti v e replisomes in a single molecule experiment in vitr o , a t concentrations similar to those found in vivo , leads to replisome destabilization and significant lagging strand gap forma tion. Ef fects of RecF on leading strand replisomes can be demonstrated in bulk experiments. Although much remains to be done , these results , along with the replisome instability seen in vivo when RecF is ov er-e xpressed, suggest that RecF could be involved not only in targeting repair to post-replication gaps but also in the replisome disengagement that results in gap creation.
When combined with the e xtensi v e literature linking RecF in some manner to RecA filament formation in gaps, the results suggest an important link between gap formation at the replication fork via lesion-skipping and the subsequent processing of those gaps by RecA protein. That link appears to be organic to RecF function, as RecF colocalization with replisomes is frequent in vivo when RecF is present at physiological concentrations ( 52 ).
How do we tie all of this to the well-established function of RecF in loading RecA protein into gaps? RecF can clearly influence RecA filament assembly if it is properly positioned ( 46 , 47 , 50 ). Howe v er, neither R ecF nor R ecFR binds at gap ends with anything like the specificity re-quired to direct these proteins to gaps in vivo . As already noted, specific RecFR binding to gap ends could be detrimental, as gaps are a feature of the replication process ( 119 ) and those arising from lagging strand DNA synthesis and mismatch repair are presumably much more abundant than lesion-containing post-replication gaps. So how could RecFR get to the location where it is needed to affect RecA filament forma tion a t lesion-containing gaps without interfering with other aspects of replication? We suggest that RecF interacts in some manner with replisomes via DnaN (and possibly DnaG) and is placed at gap ends by the replisome in the process of gap generation via lesion-skipping. In this scheme, the replisome is the lesion sensor and the RecFreplisome interaction provides an avenue for placing RecF at the end of the lesion-containing post-replication gap.
These ideas are outlined in the model in Figure 10 . Although much of the scheme is speculati v e, it builds on the growing evidence that RecF interacts with the replisome and r epr esents a first attempt to address the questions posed in the Introduction. This is also an attempt to accommodate many observations accumulated over decades as well as the current work. In particular, the model incorporates the proposed targeting role of the RecFR complex and the loading function of RecOR ( 21 , 45 , 120 , 121 ). The model also incorporates a handoff between RecF and RecR that is implied but not discussed in some earlier studies ( 50 ).
In the model of Figure 10 , RecF first interacts with an acti v e replisome through DnaN. When that fork encounters a lesion that triggers lesion-skipping (possibly with a disengagement mechanism that utilizes RecF), the replisomeassociated RecF is left behind at the gap near the 3 terminus wher e r eplication was interrupted. If the replisome undergoes some conformational change that leads to disengagement and lesion-skipping, one that does not occur during normal lagging strand DNA synthesis, this could provide a molecular signal for specific RecF positioning at the end of what then becomes a lesion-containing post-replication ga p. Ga ps generated during lagging strand DNA synthesis or mismatch repair would not be affected.
RecF dimerization and interaction with RecR strengthens the binding. The gap is likely expanded by the action of R ecJ ( 122 ). R ecA filaments, once nucleated, grow primarily in the 5 to 3 direction. If the RecFR is near the 3 gap terminus, this would position RecFR away from RecA nucleation sites within the gap, at the end where it could limit RecA filament extension beyond the gap ( 44 , 120 ). Howe v er, via looping of the intervening DNA, RecF could presumably transfer RecR to a RecO monomer interacting with SSB within the gap. A handoff of this kind could be part of a mechanism to constrain RecOR function to ga ps w her e r epair was needed. An interaction between RecF and DnaG might also facilitate a positioning of RecF at the 5 gap terminus as DNA synthesis is re-initiated following gap creation.
The best way to control RecO action is to control its access to RecR and SSB, both of which are essential to its RecA-loading function. The cellular concentrations of the RecF, RecO and RecR proteins are normally quite low. In addition, there are many cellular proteins that bind to the SSB C-terminus ( 123 , 124 ). Unlike the situation in vitro where purified proteins are used, the many SSB-binding Figure 10. Model of RecF ATPase function near the replisome. Schematic representation of RecF ATPase activity triggering the localization of RecF near the replisome. The light green circle represents the stable replisome. In the front of the replication fork the DnaB helicase (dar k b lue) unwinds the dsDNA and interacts with DnaG (dark orange). DnaG promotes RNA priming on the lagging strand. On the lagging strand, the ssDNA region intermittently formed during replication is coated by the SSB protein (light orange tetramer). The clamp loader (dark green) interacts with the two polymerase cores of the leading and lagging strand, the clamp loader also interacts with a third PolIII core that would be loaded on the next RNA primed site. Those interactions allow the integrity of the replisome. The ␤-clamp (yellow) increases the processivity of PolIII is represented in yellow. Gap formation occurs upon encounter with a lesion. RecF (pink), initially associated with the replisome, is deposited at the 3 end of the interrupted DNA strand. Stability of the bound RecF is increased by binding to RecR (purple). Gaps formed on the lagging strand have RecFR at one end of the ga p. Ga ps formed in the leading strand may have RecFR positioned at one or both ends of the ga p. Finall y, RecA is loaded onto the SSB coated DNA by the RecO (marine green) and RecR proteins at a site within the ga p, potentiall y facilitated by RecR handoff from RecFR. The authors encourage readers to compare the elements of this model with speculation offered by Kuzminov in 1999 ( 60 ).
proteins in vivo could limit RecO access to ssDN A ga ps. A handoff scheme where RecFR marked the gaps requiring repair intervention and then recruited RecO could be part of a broader regulatory process controlling RecA filament formation and its capacity to block replication forks and induce SOS.
The interactions between RecF and the replisome may not result in significant replisome instability under nor-mal cellular conditions where RecF is present at low levels. Howe v er, ov er-e xpression of RecF might lead to replisome impediment. If RecF is directly involved in gap creation, it might facilitate the intrinsic capacity of the replisome for lesion-skipping ( 7 ), using its ATPase function. RecF ov er-e xpression could thus trigger more frequent replisome disengagement with the template. Alternati v ely, the incr eased r eplisome instability noted with RecF-