Structural basis for broad anti-phage immunity by DISARM

In the evolutionary arms race against phage, bacteria have assembled a diverse arsenal of antiviral immune strategies. While the recently discovered DISARM (Defense Island System Associated with Restriction-Modification) systems can provide protection against a wide range of phage, the molecular mechanisms that underpin broad antiviral targeting but avoiding autoimmunity remain enigmatic. Here, we report cryo-EM structures of the core DISARM complex, DrmAB, both alone and in complex with an unmethylated phage DNA mimetic. These structures reveal that DrmAB core complex is autoinhibited by a trigger loop (TL) within DrmA and binding to DNA substrates containing a 5′ overhang dislodges the TL, initiating a long-range structural rearrangement for DrmAB activation. Together with structure-guided in vivo studies, our work provides insights into the mechanism of phage DNA recognition and specific activation of this widespread antiviral defense system.

T he threat of bacteriophage has driven bacteria to evolve a myriad of antiviral defense systems, capable of targeting phage at various stages of infection and replication 1,2 . The first lines of antiviral defense are typically rapid, broad, innate immune responses that include preventing phage adsorption 3 , degrading foreign DNA using restriction-modification (RM) systems 4 , or ultimately host cell destruction through abortive infection (Abi) systems 1 . Adaptive immune responses (including CRISPR-Cas systems) provide a highly-specific and long-term antiviral protection 5,6 . A number of recently discovered antiphage systems have expanded the arsenal of defense tools used by microbes in the arms race against phage [7][8][9] .
RM systems are the most highly abundant bacterial defense system, found in~75% of all bacterial genomes 10 . They often consist of DNA methyltransferase, restriction endonuclease, and target recognition modules. The vast majority rely on sequencespecific DNA methylation, which is recognized by and rapidly degraded by specific nucleases, allowing for self-versus-non-self discrimination to avoid autoimmunity. However, phage have adapted to evade such systems by either evolving to lack the restriction site sequences required for DNA cleavage, or carrying their own epigenetic modification to prevent being targeted 4,11 .
The largely uncharacterized DISARM (Defense Island System Associated with Restriction-Modification) systems are widespread in bacteria 7 . The DISARM operon typically contains a DNA methyltransferase (DrmMI and/or DrmMII that methylate adenine and cytosine, respectively) along with a helicase (DrmA), a DUF (domain of unknow function)1998-containing protein (DrmB), a phospholipase D (PLD) domain nuclease (DrmC), amongst additional auxiliary genes 7 (Fig. 1a). DUF1998 domains are common in various defense systems including Druantia 8 and Dpd 12 , suggesting a common role in anti-phage activity. While the presence of methyltransferase and nuclease genes within the DISARM operon hints that this system behaves akin to other RM systems, DrmC appears to be non-essential for DISARM activity, and DISARM can still restrict phage that lack methylation target sequences 13 . Thus, the molecular mechanisms underlying phage targeting remain unclear.
Here, we report cryo-EM structures of DrmA:DrmB (DrmAB) in the presence or absence of target DNA, revealing the arrangement of conserved RecA helicase domains relative to the DUF1998 domain. In vivo studies of structure-guided DrmAB mutants demonstrate that ATP hydrolysis, DNA binding, and DrmAB heterodimer formation are essential for phage targeting by DISARM. We observe that DrmA contains an unstructured trigger loop that partially occludes the DNA binding surface on the complex. This loop limits DNA binding for dsDNA and allows DrmAB to discriminate between targets based on DNA structure rather than DNA sequence as a mechanism of selfversus-non-self-discrimination. DNA loading induces long-range conformational changes likely associated with DISARM activation. Our results elucidate how DISARM systems can provide broad anti-phage activity, while avoiding autoimmune activation and reveal key molecular mechanisms that underpin rapid DIS-ARM activation upon phage infection.

Results
Architecture of DrmAB nucleoprotein complex. We purified the Serratia sp. SCBI (Serratia) DrmAB complex after coexpression of DrmA and DrmB in E. coli. SDS-PAGE analysis of the peak fraction from size-exclusion chromatography confirmed the presence of both DrmA and DrmB, in a~1:1 stoichiometry ( Supplementary Fig. 1). Candidate nuclease DrmC did not copurify with DrmAB when co-expressed, supporting the notion that it does not constitute part of the core DISARM complex 7 . To capture the DrmAB complex bound to a phage singlestranded DNA mimic, we incubated DrmAB with an unmethylated forked DNA substrate in the presence of ADP. DrmA-B:ADP:DNA complex formation was confirmed by native electrophoretic mobility shift assays ( Supplementary Fig. 1).
We used cryo-EM to investigate the architecture of the DrmAB:ADP:DNA complex. During data collection, we observed that the complex adopts a preferred orientation in vitreous ice, which was ameliorated through collecting additional data at a −30°tilt. Multiple rounds of classification resulted in a 3D reconstruction of the complex at a nominal resolution of 2.8 Å and focused 3D classification improved densities corresponding to DNA and the N-terminus of DrmB. Notably, we were able to separate DNA-bound from DNA-free particles through focused 3D classification, ultimately yielding 3D reconstructions of DrmAB:ADP:DNA and DrmAB:ADP at nominal resolutions of 3.3 and 3.4 Å, respectively ( Supplementary Fig. 2). Between these two structures, we determined atomic models that account for >98% of the total residues of the~220 kDa protein components, with an additional seven nucleotides of DNA in the DrmA-B:ADP:DNA complex ( Supplementary Fig. 3).
The DrmAB complex exhibits a bi-lobed architecture, resembling a partially open clam shell (Fig. 1b, c). The top lobe contains the RecA1 domain of DrmA, while the bottom lobe contains the DrmA RecA2 domain and the entirety of DrmB.
As expected for a superfamily 2 (SF2) helicase, DrmA contains tandem RecA modules (residues 600-1150) that form an active site for ATP-binding and hydrolysis and DNA loading. The N-terminal half of DrmA acts as a structural chassis, holding these two helicase motor domains in place, and forming an interface with DrmB. While RecA1 and RecA2 have high structural similarity to many other SF2 helicase domains (Zscore >14 for more than 10 other helicase structures 14 ), no other regions of DrmA were found to have significant structural homology to known protein domains. The solvent-exposed interface between the tandem RecA domains forms the nucleotide-binding site, while the bound DNA straddles these domains within the heart of the complex (Fig. 1b, c). Only four of the seven deoxynucleotides observed within our structure are solvent-exposed, with the remaining three bases deeply buried within the DrmA(RecA1,RecA2):DrmB interface.
DrmB contains a C-terminal DUF1998 module, a domain often enriched within bacterial genomic defense islands 7,15 . DUF1998 domain sits towards the back of the complex, predominantly buried within DrmB, positioned towards the interface between the two RecA domains of DrmA. Like other proteins containing DUF1998 domains, DrmB contains the predicted zinc-coordinating four cysteine motif (C559,565,581,584, Supplementary Fig. 4) 16,17 . Surprisingly, we also observed an additional four putative coordinated metal ions at highly conserved sites within DrmB ( Supplementary Fig. 4). While it is not possible to unambiguously determine the identity of these ions, based on the coordinating residues Zn 2+ is a strong candidate 18 .
This raises the possibility that DrmAB may utilize these three highly conserved metal ion coordination sites to sense changes to cellular redox potential, triggered by environmental stresses such as phage infection, as has recently been demonstrated for the type III-A CRISPR system within Serratia [19][20][21][22] . Alternatively, these clusters may play a structural role, enabling proper folding of DrmB, as is the case for the DUF1998-containing StfH protein 17 .
Multiple interaction surfaces contribute to the stabilization of the DrmAB complex, with a total buried surface area of~4000 Å 2 . Most notably, the C-terminus of DrmA (residues 1296-1319) wraps around the entirety of the N-terminus of DrmB (Fig. 1b, c), with a total buried surface area of~1850 Å 2 . Additionally, the DrmB DUF1998 domain mediates~1230 Å 2 of surface contacts with DrmA (Fig. 1d). DUF1998 is positioned towards the interface between the two RecA domains of DrmA but makes limited contacts with both (~130 and~110 Å2 buried surface area with RecA1 and RecA2, respectively). RecA2 makes additional surface contacts with the N-terminus of DrmB (482 Å 2 ). Based on this network of inter-subunit interactions, we hypothesize that DUF1998-containing DrmB may act as a modulator of DrmA helicase activity. This is supported by the observation that DUF1998 domains are frequently associated with helicase domain-containing proteins, either as adjacent genes or as C-terminal fusions 15 .
DNA binding requirements of DrmAB. The DNA-bound DrmAB complex includes a 7-mer ssDNA segment (Fig. 2a). While this is only a fragment of the DNA used for complex assembly, comprising a 19-bp duplex with a 5-nucleotide hairpin on one end and a 7-nucleotide 5′ tail and a 21-nucleotide 3′ tail on the other end, we attributed this to the flexibility of DNA that does not make direct contact with the complex. We tested the possibility of DrmAB requiring a 7-nt overhang to bind DNA by measuring the binding of DrmAB to DNA substrates with various 5′ overhang lengths. We observed binding of DrmAB to a 7-nt 5′ ssDNA overhang, and minimal binding to substrates with fewer than 7-nt overhangs ( Supplementary Fig. 1e).
DrmA has multiple interactions with the backbone of the DNA, mostly electrostatic contacts between positively charged side chains and DNA phosphate groups (Fig. 2b, c). Such nonspecific electrostatic DrmAB:DNA contacts provide the molecular basis for the previously observed broad anti-phage targeting by DISARM 7 . Since conventional RM systems require recognition of specific DNA sequence motifs for nuclease activity, they can be easily evaded through phage evolving escape mutations 11,23 . However, by lacking sequence preferences for DNA binding, DISARM can provide effective anti-phage defense in the absence of a particular restriction site.
To address the functional relevance of these interactions, we mutated various DNA-interacting residues within DrmA (K803, R1294, R659, R810) to alanine. All four mutants showed reduced anti-phage protection in vivo (Fig. 2d). Together, these data underscore the functional role of DrmA as the DNA targeting arm of the complex and provide the mechanism of targeting a wide range of phage by DISARM. Overall, the non-specific interactions between DrmA and DNA provide a structural basis for broad phage targeting by DISARM. (Fig. 2e) and DNA-free maps, both of which were at the interface between the RecA1 and RecA2 domains, the canonical ATPbinding site of SF2 helicase domains. While the RecA domains of DrmA suggest that it is a helicase, the impact of its helicase activity on anti-phage targeting is poorly understood. We mutated four ADP-interacting residues to test their impact on DIS-ARM activity. All four mutants showed reduced DISARM activity against a broad range of phage in vivo. This confirmed that in addition to DNA binding, ATP hydrolysis by DrmA is essential for DISARM activity. This is similar to type I RM systems, which encode a DEAD-box helicase domain protein that drives DNA translocation upon recognition of an unmethylated restriction site 10,24 . Akin to DrmAB, ATP-dependent DNA translocation is essential for type I RM system activity 25 . However, it is unlikely that these two anti-phage systems share a common DNA degradation mechanism, since type I RM systems translocate and cleave dsDNA, while DISARM appears to bind exclusively ssDNA (Fig. 3).
DrmA contains an unstructured trigger loop that partially occludes the DNA-binding site. Two dominant, distinct populations (~120,000 particles each) emerged during focused 3D classification; one of which had strong DNA density (DrmA-B:ADP:DNA) and one lacking observable DNA density (DrmA-B:ADP). Comparison of our two maps revealed the presence of a 55-residue loop in DrmA that is otherwise absent in the DNAbound structure (Fig. 3a, b). Due to the flexibility of this loop, we T (2) T (3) T (4) T (5) T (6) T (7) Arg810 were unable to confidently assign sequence position. We modeled this region as poly-alanine. To further validate this model, we determined a structure of DrmAB in the absence of DNA and ADP at a resolution of 3.8 Å ( Supplementary Fig. 5). Rigid-body docking of our higher-resolution DrmAB:ADP model revealed density consistent with the presence of this loop ( Supplementary  Fig. 5). Thus, the presence of this loop is mutually exclusive with DNA binding. After superposition of our two models, we were surprised to observe a severe steric clash between DNA and this unstructured loop (Fig. 3c) that occurs at the interface between the 5′ DNA end and RecA2. We hypothesized that this region of DrmA acts as a trigger loop (TL) (residues 176-232) that activates DrmAB after being dislodged by target DNA. Given that the presence of TL and DNA are mutually exclusive, we hypothesized that in the absence of single-stranded DNA, this loop may partially occlude the DNA binding site, preventing DrmAB from loading onto non-phage (i.e., host) DNA. To study its relevance in the DISARM mechanism, we created a mutant lacking this loop (DrmA(Δloop)), removing residues 181-233. Since DrmA(ΔTL) maintained its ability to co-purify with DrmB, and the DrmA(Δloop)B complex eluted at the expected volume according to size-exclusion chromatography ( Supplementary Fig. 1a), heterodimer assembly does not depend on this region of DrmA. We performed native electrophoretic mobility shift assays to compare binding of DrmAB and DrmA(ΔTL)B to fluorescentlylabeled 75-nt single-stranded DNA (Fig. 3d, e). While DrmAB bound DNA at concentrations of 1 µM or higher, the presence of smeared bands likely indicated dissociation of DNA from the complex. In contrast, DNA dissociation from DrmA(ΔTL)B bound DNA was not observed. Furthermore, a second super-shift occurred at protein concentrations of 1 µM or higher, likely corresponding to multiple copies of the complex binding to the DNA concurrently, as observed for other nucleoprotein complexes 26 .
Since TL limits DNA binding, we performed additional binding assays to determine how this loop affects DNA substrate preferences of DrmAB (Fig. 3f, g). DrmAB bound to ssDNA and  DNA with a 5′-overhang (ovh) but did not bind to dsDNA or DNA with a 3'ovh. DrmA(ΔTL)B displayed no DNA substrate preference, binding to ssDNA, dsDNA, and DNA with both a 5'ovh and 3'ovh. Thus, the TL limits structure-specific DNA binding by partially occluding the DNA-binding surface of DrmAB. While the methylation modules of the DISARM operon recognize specific sequences, DISARM targets a diverse range of phage DNA sequences 7 . We thus propose that TL functions as a specificity filter, preventing interactions with dsDNA and DNA containing 3'ovh. Since DNA would occupy a larger surface on DrmAB than TL, DNA with a 5'ovh would be an ideal substrate for efficiently competing with TL for loading onto the complex. This may function as a mechanism for conferring substrate specificity to DISARM and prevent interactions with bacterial chromosome and other self-DNA, which would likely result in deleterious autoimmune effects. By relying on DNA structural context rather than simply DNA sequence, DISARM appears to utilize an alternative mechanism for avoiding self-targeting to many other nucleic acid-based anti-phage systems 5,24 .
Methylation sensing and DNA-mediated DrmAB activation. We sought to visualize additional conformational changes within DrmAB associated with DNA loading and complex activation. Within the core DNA-binding site of DrmAB, continuous cryo-EM density is observed between the DNA nucleobase at position 3 (T (3) , Fig. 2c) and the side-chain of DrmA(V1296), likely corresponding to a van der Waals interaction (Fig. 4a). Since this residue is highly conserved (found in >98% of top 250 DrmA homologs) and is the sole interaction between DrmAB and a DNA base, we hypothesized that DrmA(V1296) may act as a sensor for DNA methylation status. Serratia DISARM operon contains both a 5-cytosine DNA methyltransferase and an N6adenine DNA methyltransferase gene which target ACAC(mA)G and YMT(mC)GAKR motifs, respectively (Fig. 1a), which may constitutively methylate the host genome to allow distinguishing self (i.e., methylated) from non-self (i.e., unmethylated) phage DNA, akin to RM systems 7,11 .
Further analysis of our structure revealed that DrmA(V1296) is in close contact with the faces of the DNA nucleobases, tightly wedged between DNA positions 3 and 4. Modeling this position as a bulky tryptophan (W) side-chain showed severe steric clashes, and this mutant rendered DISARM unable to protect against phage in our in vivo assay (Fig. 4b). We then tested the in vivo anti-phage activity of DISARM with a glycine at this position (DrmA(V1296G)), which would no longer contact DNA nucleobases. This also resulted in severely reduced DISARM activity, supporting the notion that V1296 plays a critical role in DNA recognition (Fig. 4b). We propose that since this interaction is critical to DISARM function, differences in DNA methylation status may perturb this interaction, providing a structural mechanism for enabling DISARM to differentiate between selfand non-self DNA and preventing autoimmune targeting.
To further test this hypothesis, we performed ATPase assay of DrmAB in the presence and absence of methylated and unmethylated DNAs. In the absence of DNA, DrmAB did not exhibit ATPase activity (Fig. 4b). This is expected for a SF2 helicase, where DNA translocation is typically coupled to ATP hydrolysis 27 . Robust ATPase activity is strongly stimulated by unmethylated DNA, but the ATPase rate is reduced in the presence of a methylated DNA substrate (containing three separate 5-methyl-cytosine bases, i.e., the modification provided by DrmMII 7 ). This supports the model that this multi-protein complex detects phage DNA as it is injected into bacteria or is being replicated. Additionally, recent studies demonstrate that DISARM can protect against plasmid conjugation, and that DISARM targeting is enhanced by an order of magnitude in the absence of cognate methylation target sequences 28 . This data indicates that DISARM is repressed by methylation rather than being activated by the presence of a specific signal sequence.
We then compared our DNA-bound and DNA-free DrmAB complexes by superposing our models and generating motion vectors ( Supplementary Fig. 6). Upon loading of unmethylated DNA, we observed minor (mostly < 5 Å) conformational changes of the RecA1 and RecA2 domains (Fig. 4a). Surprisingly, we observed large conformational rearrangements in the N-terminal half (NTH) of DrmB, despite a lack of any contacts with DNA. DrmB NTH shifts~10 Å towards the RecA2 domain of DrmA, tightly clamping the complex around DNA.
This long-range allosteric communication appears to be mediated by the C-terminal region of DrmA, which tightly wraps around the N-terminus of DrmB acting as a Pivot Arm (PA). While DrmA(V1296) undergoes a minor shift upon recognition of unmethylated DNA, this is propagated into a much larger conformation change in the end of the PA and DrmB NTH. Thus, conformational change upon loading unmethylated DNA activates DrmAB. Since the DUF1998 domain of DrmB acts a hinge for the conformational changes induced by the PA, we propose that DUF1998 should be renamed as a Helicase Allosteric Relay (HAR) domain.
DrmA PA is tightly interwoven with DrmB NTH, forming a myriad of contacts. These include a network of electrostatic interactions, hydrophobic contacts and π-π stacking (Fig. 4d). We also observed that a five-residue segment of DrmA PA contributes to a 5-stranded β-sheet in trans with DrmB, forming an oligonucleotide/oligosaccharide-binding (OB) fold (Fig. 4e). The β-strand from DrmA is sandwiched between strands from DrmB, further locking the PA of DrmA in place. These intimate contacts allow DrmA to allosterically communicate with DrmB, setting in motion major conformational changes upon loading of unmethylated DNA.
To test the importance of this conformational change for DISARM activity, we truncated DrmA so that it lacked the PA and tested the ability of DISARM to prevent phage replication in vivo. We found that DrmA-ΔPA had severely reduced DISARM activity against a broad range of phage, supporting the notion that the CTR-driven conformational changes of DrmAB result in DISARM activation (Fig. 4f). Based on our structural and functional data, we propose that upon displacement of the TL by ssDNA, DrmA(V1296) senses the DNA methylation status, allosterically activating DrmAB and triggering the DISARM anti-phage response.
In summary, through coupling ATPase activity to DNA nucleobase methylation status, DrmAB may be able to achieve specific targeting of unmodified non-self (i.e., phage) DNA without relying on a given DNA sequence.

Discussion
We propose a model whereby DrmA and DrmB are expressed by the host cell in an autoinhibited form in the absence of phage infection (Fig. 5). Through the constitutive expression of the core DrmAB complex, a DISARM response can be rapidly activated upon recognition of phage stimulus without necessitating transcription of either component, in accordance with previous observations 7 . In the absence of DNA, the DNA-binding channel of DrmAB is occluded by an unstructured~50 residue loop within TL. While truncation of this loop does not affect heterodimer formation, we observed that DrmA(Δloop)B no longer exhibited discrimination for 5'ovh-containg DNA substrates and could bind both 3'ovh and dsDNA. We conclude that this loop   [30][31][32][33] . Interestingly, a recent study of the human DEAD-box helicase DHX37 revealed a highly similar autoinhibition mechanism, whereby an unstructured protein loop occupied the substrate-binding channel 34 . It may emerge that similar mechanisms of autoinhibition are widespread amongst a diverse group of SF2 helicase proteins from all domains of life. Based on our data, we propose that the activating substrate for DISARM is unmodified 5'ovh DNA. Unlike many RM systems, DISARM does appear to not rely on specific sequences, enabling broad anti-phage targeting that cannot be circumvented through escape mutantion. This is further supported by in vivo functional data that show that DISARM activity is enhanced by a lack of cognate methylation target sequences 28 . Future studies are required to investigate the interplay between DrmAB and other less-conserved DISARM components (e.g., DrmD and DrmE) within DISARM, and how these subunits affect the specificity of DISARM activation.
The preference for 5'ovh ssDNA is particularly important since many dsDNA phage inject their genome in a linearized form with two sticky end 5'ovh, which subsequently become ligated during DNA circularization 35,36 , thus providing a stimulus for DISARM activation. This is supported by the previous observation that DISARM has no effect on phage adsorption but blocks phage DNA circularization 7 . If a dsDNA phage does not have such 5' ovh, then it would still likely be targeted by DISARM since it would still replicate via rolling circle DNA replication, where the leading strand is replicated in a 5′ to 3′ direction and the lagging strand is synthesized as Okazaki fragments 37 . This model explains the previous observation that DISARM is activated by the onset of phage replication, likely resulting probably in the degradation of viral DNA 7 . This data also explains how DrmAB target specificity is conferred by DNA structure rather than a specific sequence, providing an elegant mechanism for balancing broad DNA targeting with minimal autoimmune consequences. DIS-ARM therefore represents a novel paradigm in bacterial antiviral defense mechanisms that target nucleic acids. Rather than achieving specific and efficient degradation of phage DNA through recognition of specific sequences, DISARM recognizes the structural context of DNA. The core DISARM complex DrmAB is able to confer specificity of DNA loading by targeting substrates with a 5'ovh, thereby avoiding degradation of the host chromosomal DNA. This specificity is further conveyed by a sensitivity to DNA methylation status, whereby methylated DNA limits the ATPase activity of DrmAB which is essential for antiphage defense. This amounts to a two-pronged mechanism of selectivity that does not rely on a given DNA sequence, and thereby circumvents evasion via mutational escape.
While this model is consistent with our data, the class 2 DIS-ARM system has been demonstrated to confer protection against modified phage 7 . While DISARM preferentially targets unmodified DNA, methylated DNA can still support ATP hydrolysis and thus defense activation. It may be the case that DISARM activation is significantly reduced by phage DNA methylation, but the abundance of phage 5'ovh during successive cycles of replication may provide sufficient stimulus to activate DISARM. Additionally, other DISARM subunits may confer additional mechanisms to detect invading phage.
We hypothesize that once activated, DrmAB binds to, translocates, and unwinds phage DNA, blocking replication and  transcription and disrupting the phage infection cycle. Additionally, DrmAB may recruit a nuclease to trigger DNA degradation. Since the PLD-containing candidate nuclease DrmC is dispensable for DISARM activity and is often missing from annotated DISARM operons, it has remained a mystery how this is achieved 7 . However, since PLD-containing nucleases are highly abundant in bacterial genomes, we propose that DrmAB may recruit an alternative PLD-nuclease protein. While previous in vivo assays were performed in Bacillus subtilis (which encodes a 42.2% identity DrmC-like homolog), our in vivo assays were performed in Escherichia coli (which encodes a 93.4% identity DrmC-like homolog). The ubiquity and versatility of such DrmClike non-specific PLD nucleases 38-41 may make DrmC redundant, allowing DISARM to 'mix-and-match' components from other innate immune systems encoded by the host bacteria in order to achieve rapid and broad anti-phage protection. While CRISPR-Cas systems rely on RecBCD and other host DNA double-strand break repair complexes for adaptation (i.e., acquiring spacers) [42][43][44] , it is unlikely that DISARM utilizes this mechanism, since it does not require spacer acquisition of phage target sequences. Phage therapy represents a promising avenue to treat bacterial infection in an age where antibiotic resistance is widespread 45 . Since DISARM is an incredibly widespread mechanism of antiviral defense in bacteria, once we understand the fundamental mechanisms underlying DISARM activation it may be possible to develop small molecule treatments to inhibit DISARM, which could be delivered in conjunction with phage therapy to increase the effectiveness of such antibacterial treatment.

Methods
Cloning and protein expression. Genes drmA and drmB were amplified from genomic DNA of Serratia sp. SCBI using primers BN1750, BN1751, BN1060 and BN1061 (Supplementary Table 3) with Q5 DNA Polymerase (NEB: M0491S) as indicated by the manufacturer. Amplified gene drmB was cloned in plasmid 13S-S (Addgene: 48329), resulting in a N-terminal 6X His tagged to the protein product. His-tag was separated from the gene with a tobacco etch virus (TEV) protease recognition sequence. Amplified gene drmA was cloned in plasmid pACYCDuet-1 (Sigma-Aldrich: 71147), resulting in a non-tagged protein product. Cloning products were transformed in Dh5α using the heat-shock protocol 46 and confirmed by sanger sequencing (Macrogen).
Both plasmids were transformed into E. coli BL21-AI TM (Thermo Fisher) to express the DrmAB complex. Briefly, cultures were grown in LB medium at 37°C with shaking until exponential growth phase (OD 600 of 0.5). After cooling down in ice for 20 min, they were induced by addition of 0.2 % (w/v) L-arabinose and 1 mM IPTG. After overnight incubation at 20°C with shaking, cells were pelleted by centrifugation, resuspended in Buffer A (25 mM HEPES, pH 8, 600 mM NaCl, 5% glycerol, 25 mM imidazole, cOmplete™ EDTA-free Protease Inhibitor), and sonicated (10 cycles of 30 s separated by 30 s breaks, 40% amplitude). Cell lysate was pelleted down (20,000 × g, 60 min, 4°C) to remove cell debris, and supernatant was filtered with a 0.45 µm PES filter and placed on ice until proceeding with the purification.
All the mutants used here were generated by round-the-horn side-directed mutagenesis and purified as the wild-type (WT) proteins. Bacterial strains used in phage assays. DISARM system genes from Serratia sp. SCBI were cloned in plasmids (Supplementry Table 4), for both WT and mutant. Plasmids were transformed in E. coli BL21-AI. For assays, strains were cultured in LB media at 37°C, and induced with 0.1% (w/v) L-arabinose and 0.5 mM IPTG at early exponential growth phase (OD 600 of 0.2-0.3). Cultures were incubated at 37°C for 90 min and then used for phage assays. In parallel, a control strain (BL21-AI WT) was grown for all assays. When required, LB media was supplemented with antibiotics at the following final concentrations: 100 µg/mL ampicillin, 50 µg/ mL kanamycin, 50 µg/mL streptomycin, 50 µg/mL spectinomycin, 10 µg/mL chloramphenicol.
Phage strains. E. coli phages T1, T7, and Nami were used in this study. For their production, phages T1 and T7 were propagated in E. coli BL21-AI, as described previously 47 , and bacteriophage Nami in its host E. coli isolate R10256 following same procedure. Briefly, bacterial cultures at exponential growth phase (aprox. 0.4 OD 600 in a 10 mm cuvette readers) were infected with a phage lysate and incubated overnight. Then, cultures were spun down and the supernatant filtered through 0.2 µm PES membranes. When required, phages were concentrated by addition of PEG-8000 and NaCl at final concentrations of 100 mg/mL and 1 M, respectively. Following, it was incubated overnight at 4°C, centrifuged at 11,000 × g at 4°C for 60 min, and the phage-containing pellet resuspended in the desired final volume of Saline Magnesium (SM) buffer. Phage stocks were stored at 4°C before their use and titer determined as indicated below.
Phage titering. Bacterial cultures in exponential growth phase were used to titer phages stocks. For this, 100 µL of culture were mixed with 5 mL of 0.6% LBA at 45°C and poured to a LBA plate to form a bacterial layer. Ten-fold dilution of phages in SM buffer or LB media were plated on the top of the bacteria in 10 µL drops and let dry for 20 min. Plates were incubated overnight up-side down at 37°C. To determine the phage titer, the number of center of infections (plaques) were counted. E. coli BL21-AI was used for titering phages T1, T7, and Nami.
Efficiency of plating determination. To determine the efficiency of plating (EOP), phages were plated on the induced strains containing the DISARM system and compared to the plating on control strain BL21-AI. For this, 200 µL of the bacterial cultures were mixed a specific, countable number of infectious particles (50-150 PFU/plate) and 4.5 mL of 0.6% LBA prewarmed at 45°C. The mixture was then poured on top of a LBA plate to form a bacterial layer containing the infectious particles. After overnight incubation at 37°C, the EOP was calculated by dividing the number of plaques counted on each plate by the number of plaques formed in the control strain.
CryoEM sample preparation, data collection and processing. To capture DrmAB in the act of unwinding a DNA substrate, a forked DNA construct consisting of a 19-bp stem-loop structure with a 6 base 5′ overhang and a 20 base 3′ overhang was designed. This substrate was chosen because similar substrates have been used to capture unwinding intermediates of bacterial SF2 helicase complexes 48,49 . DrmAB was buffer exchanged into cryoEM sample buffer (150 mM NaCl, 25 mM HEPES pH 7.4) through repeated application of buffer to sample within a 0.5 ml spin concentrator, with a 30 kDa size cutoff. 10 µM DrmAB was incubated with 10-fold excess DNA stem loop (100 µM) and 100-fold excess ADP (1 mM). Complex formation was monitored using electrophoretic mobility shift assay (EMSA), confirming DNA binding by DrmAB ( Supplementary Fig. 1). After incubation at room temperature (~25°C) for 30 min, DrmAB:ADP:DNA was applied to glow discharged holey carbon grids (C-flat 2/2, Protochips Inc.), blotted for 2 s with a blot force of 4 and rapidly plunged into liquid ethane using an FEI Vitrobot MarkIV.
Data was collected on an FEI Titan Krios cryo-electron microscope equipped with a K3 Summit direct electron detector (Gatan, Pleasanton, CA) operating in super-resolution counting mode. Images were recorded with SerialEM v3.8 50 with a pixel size of 1.1 Å over a defocus range of −1.5 to −2.5 µm. During early stages of data collection, a preferred orientation was observed. To ameliorate this, a dataset of 6828 micrographs was collected at 30°tilt, in addition to the original 2548 micrographs collected without tilt. Movies were recorded at 13.3 electrons/pixel/ second for 6 s (80 frames) to give a total dose of 80 electrons/pixel. CTF correction, motion correction and particle picking were performed in real-time using WARP v1.09 51 , resulting in 4,669,932 particles, which were uploaded to cryoSPARC v3.2 52 (Supplementary Fig. 2).
Multiple rounds of 3D classification within cryoSPARC yielded a final set of 144,989 particles that gave a 3D reconstruction at a global resolution of 2.84 Å using non-uniform refinement 53 . However, since bound DNA had weak density, an alternative data processing strategy was implemented. Two rounds of ab initio reconstruction followed by heterogeneous refinement were performed on the 0°a nd 30°tilt datasets separately within cryoSPARC. The resulting subset of undamaged particles was combined and yielded a 3.1 Å reconstruction. These particles were then imported into Relion v3.1 54 , where masks covering the core of the complex (to improve DNA density) and the bottom of the complex (to improve the quality of DrmB density) were generated within ChimeraX v1.0 55 and Relion. These masks were then used for focused 3D classification within Relion (N = 6, T = 25). Particles within classes corresponding to DrmAB:ADP:DNA and DrmAB:ADP were then selected for Relion 3D auto-refinement and postprocessing, resulting in structures with global resolutions of 3.4 Å and 3.3 Å, respectively ( Supplementary Fig. 2). Additionally, A dataset of DrmAB alone was collected and processed as described above, yielding a final reconstruction with a resolution of 3.84 Å (Supplementary Fig. 6) To assist model building, the structures of multiple overlapping regions of DrmA and DrmB were predicted using the trRosetta server 56 . These models were initially fitted into the map as rigid bodies using ChimeraX. Once suitable fits were found, bulk flexible fitting was performed using Namdinator v20191016-5814c947 57 . Regions of the model that trRosetta failed to predict were either built de novo (in well-resolved regions of the map with local resolutions of up to 3.1 Å), or omitted (in flexible, poorly-resolved regions). Between the two structures, 98% of the total DrmAB sequence was modeled.
Real-space model improvement was performed using Isolde v1.2. 58 , and final models were subjected to real-space refinement in Phenix v1.18rc5 59 . Structural figures were prepared using ChimeraX. Structural analysis of sequence conservation was performed using the ConSurf server 60 , and visualized in ChimeraX. Motion vectors to visualize the conformational change from DrmAB:ADP to DrmAB:ADP:DNA were generated and visualized using PyMol v2.5. Schematic figure was created with BioRender.com.
For comparison of DNA binding between DrmAB and DrmA(Δloop)B, serial 2-fold dilutions from of either protein complex were incubated with 5 nM Cy5-labeled DNA. Samples were incubated and EMSAs were performed as described above.
For the overhang binding assay, a 3′-Cy5-labeled DNA substrate was hybridized to unlabeled DNAs of various lengths and binding was tested as described above.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The cryo-EM structure and associated atomic model of DrmAB-ADP have been deposited into the Electron Microscopy Data Bank and the Protein Data Bank with accession codes EMD-24938 and PDB 7S9V, respectively. The cryo-EM structure of DrmAB-ADP-DNA and associated atomic model have been deposited with accession codes EMD-24939 and PDB 7S9W, respectively. Source Data are provided with this paper. Materials and correspondence requests should be addressed to Stan J.J. Brouns (stanbrouns@gmail.com) and D.W.T. (dtaylor@utexas.edu).