Structure and mechanism of the Zorya anti-phage defence system

Abstract

Zorya is a recently identified and widely distributed bacterial immune system that protects bacteria from viral (phage) infections. Three Zorya subtypes have been identified, each containing predicted membrane-embedded ZorA–ZorB (ZorAB) complexes paired with soluble subunits that differ among Zorya subtypes, notably ZorC and ZorD in type I Zorya systems^1,2. Here we investigate the molecular basis of Zorya defence using cryo-electron microscopy, mutagenesis, fluorescence microscopy, proteomics and functional studies. We present cryo-electron microscopy structures of ZorAB and show that it shares stoichiometry and features of other 5:2 inner membrane ion-driven rotary motors. The ZorA₅B₂ complex contains a dimeric ZorB peptidoglycan-binding domain and a pentameric α-helical coiled-coil tail made of ZorA that projects approximately 70 nm into the cytoplasm. We also characterize the structure and function of the soluble Zorya components ZorC and ZorD, finding that they have DNA-binding and nuclease activity, respectively. Comprehensive functional and mutational analyses demonstrate that all Zorya components work in concert to protect bacterial cells against invading phages. We provide evidence that ZorAB operates as a proton-driven motor that becomes activated after sensing of phage invasion. Subsequently, ZorAB transfers the phage invasion signal through the ZorA cytoplasmic tail to recruit and activate the soluble ZorC and ZorD effectors, which facilitate the degradation of the phage DNA. In summary, our study elucidates the foundational mechanisms of Zorya function as an anti-phage defence system.

Main

Bacteria face frequent bacteriophage (phage) attacks and have evolved diverse defence strategies, including the restriction-modification and CRISPR–Cas systems^3,4,5. Anti-phage defence systems have key roles in modulating phage–bacteria population dynamics and have biotechnological potential⁶. Recent studies have identified many new anti-phage defence systems, with further work revealing how some systems are activated by directly sensing phage-encoded proteins, or through indirect mechanisms^1,2,7,8,9,10. As phage invasion initiates with cell envelope interactions, some defence systems might detect changes in the envelope as early infection signals. However, such defence mechanisms have not yet been identified.

Among newly discovered anti-phage defence systems, Zorya systems are widespread^1,2. There are three known Zorya types, each encoding ZorA and ZorB, which are thought to form membrane-embedded complexes related to the proton-driven flagellar stator unit MotAB^11,12,13. Moreover, each type encodes one or more cytosolic proteins of unknown function, namely ZorC and ZorD for type I Zorya systems. Most anti-phage defence systems encoding membrane-associated proteins are thought to function by disrupting or depolarizing the host membrane, leading to cell death or dormancy before phage replication completes, a mechanism called abortive infection¹⁴. A similar membrane-depolarization mechanism for ZorAB has been proposed¹, but it has not been ruled out that ZorAB could instead act as the sensor of infection.

Using single-particle cryo-electron microscopy (cryo-EM), mutagenesis, functional assays, proteomics and total internal reflection fluorescence (TIRF) microscopy, we decipher several key aspects of the Zorya defence mechanism. We identified that ZorA and ZorB form a unique 5:2 proton motive force (PMF)-driven motor complex with a long intracellular tail and propose that it acts as a phage infection sensor and signal transduction complex. After phage perturbation of the cell envelope, the peptidoglycan-binding domain (PGBD) of ZorB anchors the complex to the cell wall, and proton flow drives ZorA and its tail to rotate around ZorB. This rotation induces recruitment of the soluble effectors ZorC and ZorD, which have DNA-binding and nuclease activities, leading to the local degradation of invading phage DNA to facilitate direct (non-abortive) defence.

Zorya protects through a direct mechanism

Type I Zorya systems are widely distributed across Gram-negative phyla (Extended Data Fig. 1a–c). We therefore cloned the complete Escherichia coli NCTC 9026 type I Zorya system (EcZorI), including its native promoter, into a low-copy plasmid and used a heterologous E. coli strain¹⁵ to examine anti-phage defence. The EcZorI system provided anti-phage activity against a diverse range of phages, but some phage families tested were unaffected (Fig. 1a,b). EcZorI did not impair the adsorption of phages to host cells (Fig. 1c) but acted subsequently to prevent phage replication and burst (Fig. 1d). However, EcZorI did not defend against plasmids introduced by conjugation or transformation (Extended Data Fig. 1d,e), suggesting that some aspect of phage infection, other than the mere introduction of foreign DNA into the cell, triggers Zorya activity.

Fig. 1: Zorya has broad activity against phages through a direct immunity mechanism.

a, Schematic of the EcZorI operon. b, EcZorI defence against diverse E. coli phages, determined using efficiency of plaquing (EOP) assays. AAI, the average amino acid identity between proteins encoded by each phage (providing an estimate of the relatedness between phages). c, Adsorption of phage Bas24 onto E. coli cells possessing or lacking EcZorI. d, One-step phage growth curve for phage Bas24 infection of E. coli, with or without EcZorI, normalized to the plaque-forming units (PFU) per ml at the initial timepoint. e, Infection time courses for liquid cultures of E. coli, with and without EcZorI, infected at different MOIs of phage Bas24. ϕ, phage. f, Phage titres at the end timepoint for each sample from the infection time courses in e, measured as PFU per ml on indicator lawns of E. coli either without (control) or with EcZorI. The limit of detection (LOD) is shown by dotted lines. g, The survival of E. coli cells, lacking or possessing EcZorI, that were infected at an MOI of 5 with Bas24. h, Time-lapse, phase-contrast microscopy of E. coli cells with and without EcZorI infected with Bas24 at an MOI of 5. i, Quantification of the time-lapse microscopy in h, displaying the measured cell area relative to the initial timepoint. For b–g, data are the mean of at least three biological replicates (datapoints indicate replicates) and error bars (c and d) or shaded regions (e) represent the standard s.e.m. For i, data are mean ± s.d., derived from independent biological triplicates.

Source Data

For population-level defence by EcZorI in liquid cultures infected at different multiplicities of infection (MOIs), each phage tested affected the control populations to differing extents; however, population growth in the EcZorI samples was generally unaffected at low (<0.1) MOIs and, in some cases, also at high (>1) MOIs (Fig. 1e and Extended Data Fig. 1f,g). Importantly, the growth kinetics at early timepoints did not reveal any premature host population collapse or delayed growth for cells expressing EcZorI compared with the negative controls (Extended Data Fig. 1f). EcZorI also reduced the levels of phages detectable at the end timepoints in most infected cultures (Fig. 1f and Extended Data Fig. 1h). For a single synchronized round of Bas24 infection (at MOI 5), EcZorI increased the survival rate of infected cells compared with the control populations (Fig. 1g). Together, these results indicate that EcZorI acts through a direct rather than abortive infection mechanism. This finding was confirmed using single-cell time-lapse microscopy, where cells expressing EcZorI continued to replicate after exposure to Bas24, whereas cells lacking EcZorI lysed within 60 min (Fig. 1h,i and Supplementary Videos 1 and 2).

Zorya contains a ZorA₅B₂ complex

To investigate how ZorA and ZorB support direct defence by Zorya, we purified recombinantly expressed EcZorAB complex from cell membranes using lauryl maltose neopentyl glycol (LMNG) detergent (Fig. 2a and Extended Data Fig. 2a). Visualized by negative-stain EM, EcZorAB contains a head domain attached to a long tail, measuring approximately 700 Å (Fig. 2b). We then resolved the EcZorAB cryo-EM structure to an overall resolution of 2.7 Å, revealing an oligomeric assembly of five ZorA and two ZorB subunits (Fig. 2c–f, Extended Data Fig. 2b–f and Extended Data Table 1). The 5:2 stoichiometry is supported by quantitative mass spectrometry (MS) analyses of EcZorI-expressing cells and conserved with the flagellar stator unit MotAB complex (MotA₅B₂) and similar rotary motors^16,17 (Extended Data Fig. 2g,h and Supplementary Table 1). Overall, EcZorAB comprises four structural layers: a predicted PGBD (ZorB Thr47–Leu246), transmembrane domain (TMD), membrane-proximal cytoplasmic domain (MPCD; spanning ZorA residues Gly48–Leu127 and Lys207–Ser222) and a tail-like structure formed by the ZorA C-terminal region (ZorA Gly223–Thr729) extending into the cytoplasm (Fig. 2a,c,e).

Fig. 2: Cryo-EM analysis of EcZorAB and its architecture.

a, Schematic of EcZorA and EcZorB. b, Negative-stain EM image of purified EcZorAB particles. Scale bar, 1,000 Å. c, Representative high-resolution two-dimensional classes of EcZorAB images from cryo-EM. Domain architectures of the EcZorAB complex are shown. Scale bar, 100 Å. d, Cryo-EM map of EcZorAB. Five ZorA subunits (purple, salmon, light green, tan and coral) surround two ZorB subunits (white and dark grey) viewed from the plane of the membrane. Membrane-bound lipids are shown in yellow. The detergent micelle is shown as a translucent surface representation in cyan. The dashed lines show inner membrane boundaries. Two cross-section views of the EcZorAB TMD and tail are shown. e, Cross-section view of the EM density map from the plane of the membrane. f, Ribbon model representation of EcZorAB, with two cross-section views of the model shown. g, Composite model of the EcZorAB whole complex, represented as a surface model. The radius of the ZorA tail is indicated. CP, cytoplasm; H, helix; IM, inner membrane; PP, periplasm. Images in b are representative of at least three replicates.

The periplasmic domain exhibited flexibility relative to the TMD; accordingly local refinement was used to improve the resolution to 3.5 Å, resolving a dimerized ZorB PGBD (Extended Data Fig. 2i–k). The flexibility of the ZorA tail prevented its complete three-dimensional reconstruction, with our cryo-EM map providing density for only the first 56 tail residues, despite the negative-stain EM and MS analyses confirming that the purified complex contained intact full-length ZorA (Fig. 2b–f). Secondary-structure prediction revealed a preference for the tail to adopt α-helical structures (Extended Data Fig. 3a), suggesting that the remaining ZorA tail probably continues the experimentally observed coiled-coil with a right-handed super-helical twist. On the basis of these observations, we constructed an idealized full-length ZorAB model in which the ZorA tail forms a helical bundle projecting 70 nm into the cytoplasm, with a helical pitch of 328 Å and a radius of 9.0 Å (Fig. 2g and Extended Data Fig. 3c–e).

ZorA₅B₂ is a PG-binding H⁺-driven motor

On the periplasmic side, the C-terminal PGBDs of the ZorB subunits form a homodimer, with the dimerization interface composed of α3 and β5 from each monomer, driven mainly by van der Waals forces and electrostatic interactions. Furthermore, a C-terminal loop of ZorB caps the side of the dimerization interface (Fig. 3a,b and Extended Data Fig. 4a,b). Each monomer contains two disulfide bridges, potentially contributing to the stability and rigidity of this domain: bridges connect α1 to the β1–β2 loop and α3 to the C-terminal loop (Extended Data Fig. 4b,c). The overall ZorB dimer structure resembles that of the periplasmic domain of MotB of the flagellar stator unit and of other peptidoglycan (PG)-binding proteins^18,19,20 (Extended Data Fig. 4b,d). MotAB is kept in an inactive state by the MotB ‘plug’ regions, which are connected through a long linker to the PGBD and inhibit ion flux and rotation of MotA around MotB (Extended Data Fig. 4e,i–m). Only after incorporation of MotAB into the flagellar motor, the MotB plug is released and the PGBDs dimerize to enable PG binding¹³. By contrast, the ZorB PGBDs are already dimerized, and the PGBDs are fused without a linker to the ZorB transmembrane (TM) helices through α1 (Fig. 3a). Mutations of key residues predicted to be involved in the ZorB PGBD dimer interface, PG binding (Tyr151, Asn152, Leu155 and Arg159) and the disulfide bridges abolished Zorya-mediated phage defence, as did truncations of the C-terminal loop region (Fig. 3h and Extended Data Figs. 2k and 6a). In vitro pull-down assays demonstrated that purified ZorAB and purified ZorB PGBD bind to PG, whereas corresponding PG-binding-site mutants displayed reduced binding (Extended Data Fig. 4f–h,q). The ZorB TM helices are asymmetrically surrounded by five ZorA subunits, each containing three TM helices (TM1–TM3). ZorA TM2 and TM3 are lined directly against the ZorB TM helices, while TM1 is peripheral and faces the lipid bilayer. Lipid densities are observed around the ZorAB TMD that probably stabilize the TMD (Fig. 2d and Extended Data Fig. 2l). The ZorAB TMD is structurally related to that of 5:2 ion-driven prokaryotic rotary motors, including possessing the universally conserved and mechanistically essential aspartate residue, Asp26 in ZorB. One ZorB Asp26 is engaged with ZorA chain 3 (ZorA(3)) through contact with TM2 Thr147 and TM3 Ser184 and the other Asp26 is unengaged and points towards a lumen enclosed by the ZorA MPCD (Fig. 3a,c). The interaction modes of these two Asp26 are the same as those in the inactive state of MotAB¹¹, suggesting a similar conformational state and rotary mechanism (Extended Data Fig. 4i–m). Despite the lack of a MotB-like ‘plug’ in ZorB, there are several features potentially blocking the rotation of ZorA around ZorB in this state: ZorA(2) Pro163 induces a kink in ZorB(1) α1 near residue Val46, and two salt bridges, ZorA(2) Glu164–ZorB(1) Arg49 and ZorA(5) Glu164–ZorB(2) Arg52, and several polar interactions are located at the ZorAB periplasmic assembly interface (Fig. 3a,d). Replacing ZorB residues 46–52 with a GGGSGGS linker (to abolish the predicted rotational blockages) and then generating a cryo-EM reconstruction revealed that, for this mutant, the densities for the ZorB TM helices could not be resolved and the ZorB PGBD density is poor, suggesting that ZorA is free to rotate around ZorB (through Brownian motion) in the mutant (Extended Data Fig. 4n–p,r,s).

Fig. 3: ZorAB is a PG-binding, proton-driven motor.

a, EcZorAB viewed from the plane of the membrane, with ZorB shown as ribbons (black and white) and ZorA shown as a translucent surface. The distance between the inner membrane and PG layer in E. coli is approximately 90 Å (ref. ³²). The cysteine residues from the two disulfide bridges in the ZorB PGBD are indicated and shown as spheres. The Asp26 residues from both ZorB TM helices are indicated and shown. b, Top view of the ZorB PGBD. c, Cross-section view of the ZorAB TMD, showing ZorB Asp26 and the surrounding residues. d, Magnified view of the interactions between ZorB and ZorA at the domain assembly interface in the periplasmic space. e, The Ca²⁺-binding site. EM densities are only overlapped on Ca²⁺ ion, and the two water molecules. f, Magnified view of the interactions between the ZorB N terminus and the ZorA tail. g, Ion-translocation pathway (semitransparent surface representation in light blue) in ZorAB. Residues along the ion-permeation pathway and from the ion-selectivity filter are shown. Each asterisk indicates residues or structural elements from the neighbouring ZorA subunit. h, The effects of ZorA and ZorB mutations on EcZorI-mediated anti-phage defence against Bas24, as measured using EOP assays. Data are the mean of at least three replicates (datapoints indicate replicates) and are normalized to the control samples lacking EcZorI. ZorB(46–52>GGGSGGS), replacement of ZorB residues 46–52 with a GGGSGGS linker. Data for additional phages are provided in Extended Data Fig. 6a. i, Time-lapse phase-contrast microscopy analysis of E. coli cells expressing empty vector control or EcZorI with or without exposed to Bas24 at an MOI of 5 in the presence or absence of 30 µM CCCP. j, Quantification of the time-lapse microscopy images in i, displaying the measured cell area relative to the initial timepoint images. For j, data are mean ± s.d., derived from independent biological triplicates.

On the cytoplasmic side, the ZorA TMD and MPCD are connected by TM1 and TM3, the intracellular segments of which are joined by three vertical helices (H1–H3) and a β-hairpin motif. H3 is less ordered owing to the presence of two proline residues, Pro126 and Pro136 (Extended Data Fig. 5a,b). We found five strong spherical densities in the ZorA MPCD, each coordinated by the mainchain carboxylate groups of Asp217 and Tyr220 from the end of TM3, and the side chains of Glu86 and Glu89 from the adjacent subunit, as well as two well-resolved water molecules. On the basis of the strongly negative electrostatic environment and the surrounding coordinating residues, we assigned these densities to Ca²⁺, bridging the MPCD of two adjacent ZorA subunits and linking ZorA TM3 to its intracellular helix (Fig. 3a,e and Extended Data Fig. 5c). We also observed a water-filled ion-permeation pathway connecting the periplasmic space through the unengaged ZorB Asp26 to the cytosol (Fig. 3g). On the periplasmic side, a cavity lined by several negatively charged residues probably attracts incoming ions. Moving towards the cytoplasmic side, ZorA residues Thr147 and Ser184 resemble an ion-selectivity filter²¹ that controls ion access from the periplasm to ZorB Asp26 (Fig. 3c,g). The absence of the additional polar residues in the ion-selectivity filter that is strictly required for sodium coordination²¹ indicates that ZorAB is probably a proton-driven motor (Extended Data Fig. 4i,k,m). The pathway extends from ZorB Asp26 in the direction of the cytoplasm to the inner lumen encircled by the ZorA MPCD, where we found a highly hydrated lateral portal that could facilitate ion exit (Fig. 3g).

We next mutated residues along the ion-permeation pathway to probe their role in Zorya defence. ZorB Asp26 is essential for all models of ion translocation and motor rotation and its mutation to asparagine abolishes Zorya defence. In the ion-selectivity filter, mutation of ZorA Thr147 or Ser184 to alanine did not impair Zorya activity against Bas24 but did against other phages, whereas the double-mutant ZorA(T147A/S184A) is non-functional against all of the phages tested. Alanine substitution of ZorA Pro136, which creates a kink in the ZorA MPCD H3 helix, resulted in increased defence activity against some phages. Furthermore, increasing the side-chain size and rigidity of Ile144, near ZorB Asp26, which would sterically hinder ZorA from rotating around ZorB, leads to non-functional Zorya (Fig. 3c,h and Extended Data Fig. 6a,c–e). We further confirmed the necessity of the PMF-driven ZorAB motor function for Zorya-mediated phage protection by performing single-cell time-lapse microscopy in the presence or absence of the PMF-dissipating protonophore carbonyl cyanide m-chlorophenyl hydrazone (CCCP). The addition of CCCP did not prevent the growth of cells expressing either EcZorI or the empty vector control in the absence of phage but did impair EcZorI-mediated protection against Bas24 (Fig. 3i,j and Supplementary Videos 3–6). These observations support the idea that PMF-driven rotation of ZorA around ZorB is essential for Zorya anti-phage defence.

ZorAB tail controls anti-phage defence

One of the most notable features of the ZorAB complex is its long tail-like structure (Fig. 2b,g). Within the ZorA MPCD, ZorB N-terminal residues Met1 and Phe2 intertwine and hydrophobically block the entrance of the tail (Fig. 3a,f). On the outside of the tail, residue Arg108 from the β-hairpin motif forms a salt bridge with Glu227, and His92 makes electrostatic contact with the hydroxyl group of Tyr228 (Extended Data Fig. 5c). Disrupting these interactions leads to loss of defence against Bas24 (Fig. 3h). Inside the tail, Leu250, Leu254, Leu258 and Leu261 from each ZorA subunit make up continuous hydrophobic pentameric rings. Moreover, we observed an extra density along the tail central axis in this region, which is best modelled as a fatty acid, consistent with a predicted lipid-binding site²² (Extended Data Fig. 5d,f). Mutants targeting this hydrophobic motif abolished Zorya defence (Fig. 3h) and we observed that, although the tail bundle appeared assembled in purified ZorA(L250G/L254G/L258G/L261G)–ZorB and ZorA(L250N/L254N/L258N/L261N)–ZorB mutants, the TMD domain was perturbed (Extended Data Fig. 5g,h), suggesting that the ZorA tail influences correct ZorAB TMD assembly. To further test this, we deleted the entire tail (ZorA(∆223–729)) and this mutant also abolished ZorAB TMD motor formation (Extended Data Fig. 5i). Given that the tail structure protrudes into the cytoplasm and is surrounded by aqueous solution, hydrophobic interactions inside the tail seem to be the primary driving force for tail assembly, and it is unlikely that the tail conducts ions or other small soluble molecules (Extended Data Fig. 3d,e). Notably, part of the ZorA tail (residues 540–729) shows homology with the core signalling unit of the bacterial chemosensory array (Extended Data Fig. 3b), which contains a long intracellular helical bundle that is responsible for transferring signal from the extracellular environment into the cell and regulates the activities of the subsequent effectors²³. Sequence analyses further revealed that long ZorA tails are present in all Zorya subtypes, suggesting that the tail length is functionally important (Extended Data Fig. 6b).

Deletion of any of the Zorya genes results in the loss of anti-phage defence, emphasizing that the complete function of the Zorya system requires communication between the membrane-anchored ZorAB complex and cytosolic soluble proteins (Figs. 3h and  4c,i). Considering the motor-like structural features of the ZorAB TMD and its long cytoplasmic tail, we speculated that the ZorA tail is responsible for transmitting a signal derived from the activated ZorAB motor to the cytosolic proteins ZorC and ZorD. To test whether the length of the ZorA tail is important for Zorya function, we made four ZorA tail truncations: deleting the beginning (ZorA(∆223–482)), middle (ZorA(∆359–592)) and tip (ZorA(∆483–729)) of the tail as well as a combination of deleting the beginning and tip (ZorA(∆223–343/∆449–729)) (Extended Data Fig. 5a). Purification and cryo-EM analyses of ZorA(∆359–592) ZorB and ZorA(∆435–729) ZorB confirmed that the deletions resulted in shorter tails but did not impair ZorAB TMD motor assembly (Extended Data Fig. 5j–m and Extended Data Table 1). However, all tail truncations abolished Zorya defence (Fig. 3h and Extended Data Fig. 6a,c,d). Mutating the Ca²⁺-binding site (E86A and E89A) also resulted in the loss of Zorya function (Fig. 3h and Extended Data Fig. 6a) and caused conformational changes in the ZorA MPCD, including to the linker between TM3 and the ZorA tail helix (Extended Data Fig. 5e,n). Thus, inactivation of the Ca²⁺-binding sites probably disrupts the connection between the ZorAB TMD motor and the tail. These data indicate that ZorAB tail integrity and its motor connection are essential for Zorya anti-phage activity, supporting the role of the ZorA tail in communicating with the cytosolic effector proteins.

Fig. 4: Structural and functional characterization of ZorC and ZorD.

a, Ribbon model representation of ZorC. Residues from Arg58 to Pro478 were modelled based on EM density. b, Details of the ZorC EH signature motif. c, The effects of ZorC mutations on EcZorI-mediated anti-phage defence, as measured using EOP assays. d, In vitro interaction of EcZorC with 200 nM dsDNA (18 bp, 50% GC content, 5′ FAM-labelled dsDNA; the sequence is shown in Extended Data Fig. 7h), ZorC concentrations were as follows for lanes 1, 3, 4, 5, 6 and 7, respectively: 2,500, 100, 250, 500, 1,000 and 2,500 nM. e, The effects of ZorC mutations on dsDNA-binding activity. All reactions were made to a final concentration of 100 nM of dsDNA and 2,000 nM of protein. f, Representatives of high-resolution two-dimensional classes of apo EcZorC and EcZorC–dsDNA complex images from cryo-EM. Scale bars, 100 Å. g, Ribbon model representation of EcZorD in a complex with ATP-γ-S, with the bound ATP-γ-S shown as a sphere representation. h, Details of ATP-γ-S-binding sites. The backbone of the DEAQ box motif (ZorD residues 730–733) is coloured in magenta. Conserved negatively charged residues surrounding the DEAQ box motif are shown. i, The effects of ZorD mutations on EcZorI-mediated anti-phage defence, as measured using EOP assays. ΔNTD represents ZorD(Δ1–502) and ΔCTD represents ZorD(Δ503–1080). For c and i, data are the mean of at least three replicates (datapoints indicate replicates) and are normalized to the control samples lacking EcZorI. Data for additional phages are provided in Extended Data Fig. 6a. j, ZorD CTD degrades linear plasmid DNA. k, ZorD CTD degrades phage Bas24 genomic DNA (gDNA). Data for additional phages are provided in Extended Data Fig. 8e. l, Time-lapse montage of SYTOX-Orange-labelled Bas24 infections. The arrows indicate phage particles that appear to adsorb and inject their DNA. Scale bars, 4 µm. m, Schematic of the apparent transfer of labelled phage DNA from the capsid to inside the cell. n, Quantification of intracellular fluorescence levels over time in individual E. coli cells, comparing the infection dynamics in Zorya-deficient cells and EcZorI-expressing cells (data from l, plus four additional replicates). The dotted points indicate cell lysis of E. coli cells lacking EcZorI. The bold lines represent the mean estimated from a linear regression analysis. Images in d, e, j and k are representatives of at least three replicates. CTD, C-terminal domain; NTD, N-terminal domain; WT, wild type.

Source Data

ZorC and ZorD DNA interactions enable defence

To better understand the roles of ZorC and ZorD in anti-phage defence, we next obtained their structures and investigated their biological roles. ZorC possesses a domain containing an EH signature motif (Glu400, His443) with unknown function. We determined the cryo-EM structure of EcZorC to an anisotropic resolution of 3.7 Å, with sufficient density to model residues Arg58–Pro478 (Fig. 4a, Extended Data Fig. 7a–e and Extended Data Table 1). EcZorC consists of a core domain that connects through a long linker to a C-terminal globular domain, the density of which is blurred, and de novo model building was not possible (Extended Data Fig. 7b). In the core domain, the EH signature motif Glu400 and His443, together with Asp332, Arg335, Trp339 and Trp458, form an electrostatic network (Fig. 4b). Moreover, in an AlphaFold3-predicted²⁴ full-length ZorC model, the N-terminal region (residues Met1–Glu48) was modelled with low confidence as two hydrophilic helices that extend from the core domain beyond the density observed in the cryo-EM map (Extended Data Fig. 7b,f). Deletion of the N-terminal helices or the C-terminal globular domain results in loss of Zorya function (Fig. 4c). The ZorC surface contains several patches of net positive charge, including the region containing the EH signature motif (Extended Data Fig. 7g), suggesting that ZorC might interact with nucleic acids. Using electrophoretic mobility shift assays (EMSAs), we observed that ZorC can bind to double-stranded DNA (dsDNA) in a sequence-independent manner (Fig. 4d and Extended Data Fig. 7h). Both the EH signature motif and C-terminal globular domain are required for ZorC DNA-binding activity (Fig. 4e). We further confirmed the ZorC–DNA binding by obtaining a cryo-EM dataset of ZorC in complex with dsDNA. Although we did not obtain a high-resolution reconstruction, the 2D classes clearly indicate that the EH-motif-containing domain interacts with DNA (Fig. 4f and Extended Data Fig. 7i). Furthermore, ZorC E400A or H443A mutations abolished Zorya function, indicating that ZorC DNA binding is indispensable for Zorya defence (Fig. 4c and Extended Data Fig. 6a).

EcZorD contains a predicted Snf2-related domain at its C terminus and such domains are known to use ATP hydrolysis to bind to or remodel DNA²⁵. We therefore determined the structure of EcZorD in the absence and presence of a slowly hydrolysable ATP analogue, ATP-γ-S (Fig. 4g, Extended Data Fig. 8a–d and Extended Data Table 1). The EcZorD N-terminal domain (residues Met1–Asn502) interacts directly with its C-terminal domain (residues Asp503–Ala1080), forming a toroid-shaped molecule. ATP-γ-S is bound within a cleft near the hallmark DEAQ box motif (ZorD residues Asp730–Gln733), surrounded by many conserved negatively charged residues (Fig. 4h). Mutations in both the ATP-binding site (D730A/E731A) and those conserved negatively charged residues resulted in loss of Zorya function (Fig. 4i). We next assessed ZorD DNA-targeting activity by incubating purified ZorD with plasmid DNA in vitro. Full-length ZorD was unable to degrade DNA; however, the ZorD C-terminal domain exhibited nuclease activity and rapidly degraded both plasmid DNA and phage genomic DNA (Fig. 4j,k and Extended Data Fig. 8e). Mutating the DEAQ box motif (D730A, E731A) or a glutamate (Glu651) recognizing the ATP ribose group completely abolished the nuclease activity of the ZorD C-terminal domain (Fig. 4j). Moreover, the presence of ZorC did not activate the autoinhibited ZorD nuclease activity or inhibit the activity of the ZorD C-terminal nuclease domain (Extended Data Fig. 8f,g). This aligns with the finding that ZorC and/or ZorD alone cannot protect against phage infection without ZorAB. Our results suggest that ZorD has nuclease activity and that its full-length form is autoinhibited, probably becoming active once defence is triggered, presumably through a conformational change consistent with AlphaFold3 predictions (Extended Data Fig. 8h–m).

To directly measure ZorD nuclease activity in vivo, we attempted to use the parS-ParB system to track phage DNA within infected cells²⁶. While the system was not functional with Bas24, we were able to establish activity with Bas54, against which Zorya also confers protection (Fig. 1b). Using time-lapse microscopy to track the presence of phage DNA during Bas54::parS infection, we observed Bas54::parS DNA within infected cells (detected as fluorescent spots of ParB–mScarlet bound to parS loci) in the absence but not presence of Zorya (Extended Data Fig. 7j and Supplementary Video 7). This supports our model for Zorya-mediated phage DNA degradation, but could also be explained if Zorya blocked injection of phage DNA. To exclude this possibility, we labelled Bas24 phage DNA with a fluorescent dye (SYTOX Orange) and used time-lapse imaging to track phage adsorption, DNA injection and subsequent fate of the injected DNA and infected cells. In the absence of Zorya, we observed individual phage particles adsorbing and injecting their DNA, apparent as transfer of fluorescence from the phage capsid to an accumulation of intracellular fluorescence. In the presence of Zorya, DNA injection still occurred (the fluorescence from the adsorbed phage particles decreased over time, consistent with the rate of DNA injection in the absence of Zorya), but we detected no intracellular fluorescence accumulation, suggesting that the injected phage DNA was rapidly degraded and that the cells were protected from lysis (Fig. 4l–n and Supplementary Video 8). These experiments provide further evidence supporting phage DNA degradation by the EcZorI system.

ZorAB recruits ZorC/D during phage invasion

We next examined how ZorAB, ZorC and ZorD coordinate during phage infection. We examined whether zorC and zorD from a Pseudomonas aeruginosa type I Zorya system could complement corresponding deletions of EcZorI genes, but this was not the case, suggesting that direct interactions occur between ZorAB and either or both ZorC or ZorD (Extended Data Fig. 9a,b). Moreover, quantitative MS analysis of EcZorI-expressing cells implied an approximate 1:1:1 ZorA₅B₂:ZorC:ZorD stoichiometry (Extended Data Fig. 2g,h). We next used TIRF microscopy to examine the subcellular distributions of functional mNeonGreen (mNG) or HaloTag (HT) fusions to ZorB, ZorC and ZorD (Extended Data Fig. 9c,d). In the presence and absence of phage, ZorB–mNG and ZorB–HT formed distinct membrane-associated foci, while expression of mNG alone from the EcZorI promoter resulted in uniform, cytoplasmic fluorescence independent of phage (Fig. 5a,c and Extended Data Fig. 9e,f). We observed a slight but significant increase in ZorB–mNG foci in phage-infected bacteria, which was independent of ZorAB motor function, as the non-rotating ZorB(D26N) mutant showed a similar increase (Fig. 5a,b). For both mNG–ZorC and ZorD–mNG, we observed a significant increase in membrane-associated focus formation after phage infection compared with the non-phage control, suggesting that the cytosolic effector proteins ZorC and ZorD are recruited to an activated EcZorI system (Fig. 5c,d). This result is reinforced by a positive correlation between the number of ZorD foci and the phage MOI (Extended Data Fig. 9h,i). We next used dual-tagged constructs to investigate co-localization of ZorB–HT with either mNG-tagged ZorC or ZorD (Extended Data Fig. 9c). In the absence of phage, ZorB and ZorC or ZorD foci rarely co-localized. However, after Bas24 infection, co-localization of ZorC–mNG or ZorD–mNG with ZorB–HT occurred in approximately 20% or 30% of cells, respectively (Fig. 5e–h). As TIRF microscopy visualizes only about a quarter of the cell depth²⁷, the observed co-localization is probably an under-representation of the ZorC/D recruitment frequency. No ZorD co-localization was detected in the non-functional motor ZorB(D26N) mutant or in a ZorA tail-tip-deletion mutant (Fig. 5h). Overall, these data suggest that the cytosolic effector proteins ZorC and ZorD are recruited to phage-activated ZorAB complexes, and that both the rotary function of ZorAB and the cytoplasmic tail of ZorA are required to transmit the phage infection signal and recruit the cytosolic nuclease to activated ZorAB.

Fig. 5: Subcellular distributions and co-localization of Zorya components with a proposed model.

a, Exemplary denoised TIRF images of the subcellular distributions of WT ZorB and ZorB(D26N) fused to HT with or without Bas24. Scale bars, 1 µm. b, Comparison of the detected maxima of the ZorAB complex foci between the untreated or Bas24-exposed (MOI, 5; 30 min) conditions. n > 250 cells, 3 replicates. P = 0.022 (WT) and P = 0.027 (D26N). c, Exemplary denoised TIRF images of the subcellular distributions of mNG-tagged ZorC and ZorD, with or without Bas24; mNG was fused to either the ZorC N terminus (mNG–ZorC) or ZorD C terminus (ZorD–mNG). Scale bars, 1 µm. d, Comparison of the detected maxima of the ZorC and ZorD foci between the untreated and Bas24-exposed (MOI 5, 30 min) conditions. n > 250 cells. P = 0.004 and P = 0.04. e, Exemplary denoised TIRF images of co-localization of mNG–ZorC with ZorB(WT)–HT with or without Bas24. Scale bars, 1 µm. The white arrows highlight co-localization. f, Co-localization analysis of ZorB(WT)–HT or ZorB(D26N)–HT with ZorC–mNG with or without Bas24 (MOI, 5; 30 min). P = 0.0002 (left), P = 0.0042 (right). g, Exemplary denoised TIRF images of co-localization of ZorD–mNG with ZorB–HT, with or without Bas24. Scale bars, 1 µm. h, Co-localization analysis of ZorB(WT)–HT or mutants (ZorB(D26N)–HT and ZorA(∆483–739) (tail-tip deletion)) with ZorD–mNG, with or without Bas24 (MOI, 5; 30 min). ****P < 0.0001. i–k, The proposed Zorya defence model. OM, outer membrane. i, Inactive ZorAB diffuses laterally within the inner membrane. j, Inactive ZorAB detects cell envelope perturbation during phage infection and ZorB PGBDs anchor to PG. Ion translocation through ZorAB triggers ZorA and its tail to rotate around ZorB. k, The ZorAB motor signal is transferred through the ZorA tail, which recruits and/or activates ZorC and ZorD. ZorC and ZorD bind to and degrade phage DNA, preventing phage replication. Datapoints represent the mean focus counts for each of three replicates and the shaded bars represent the mean between replicates. For b and d, data are mean values, with the whiskers representing the minimum to maximum values. Statistical analysis was performed using unpaired t-tests (b,d,f) or two-way ANOVA (h)³³. ZorB(WT)–HT and ZorB(D26N)–HT expression data are provided in Extended Data Fig. 6e.

Source Data

Discussion

Here we show that an E. coli type I Zorya system exhibits defence activity against phylogenetically diverse phages through a direct immunity mechanism but not against bacterial conjugation or plasmid transformation. ZorA and ZorB form an inner-membrane-integrated ZorA₅B₂ proton-driven rotary motor complex with a long, intracellular tail structure. We propose that the ZorAB complex acts as a sensor to detect phage infection and transmits the invasion signal through rotation of the ZorA tail to recruit and activate the effectors ZorC and ZorD, which bind to and degrade invading phage DNA (Fig. 5i–k). Our model and data refute a previous hypothesis that ZorC and ZorD are involved in sensing and inactivation of phage DNA, with ZorAB acting as a proton channel to depolarize the membrane potential and induce cell death if the initial ZorC/D protection failed¹. Instead, we found that the defence is direct, does not induce cell dormancy or death, and that ZorC and ZorD alone (without ZorAB) do not provide protection from phage infection.

We propose that ZorAB senses phage-induced perturbations that reduce the distance between the inner membrane and the PG layer, enabling ZorB to anchor to PG. PG binding and ion flow through the TMD are essential for ZorAB and the flagellar stator unit MotAB to function as rotary motors. However, the short distance of the ZorB PGBD to the membrane motor unit (due to the absence of a flexible linker typical for MotB) means that, normally, the PG layer is too distant for ZorB to reach (Fig. 3a and Extended Data Fig. 4e). We therefore propose that ZorAB complexes are usually inactive and free to diffuse laterally, as observed for MotAB complexes before flagellar incorporation^13,28. Perturbation of the PG layer or an increase in the local curvature of the inner membrane (which is known to occur during breaching of the cell envelope by some phages)²⁹, or other phage-induced cell envelope changes that reduce the inner-membrane–PG-layer distance (such as localized PG degradation generating ‘frayed’ edges)³⁰, would enable the ZorB PGBD to contact the PG layer, allowing binding and subsequent activation of ZorAB and recruitment and/or activation of effectors (ZorC and ZorD) to clear the phage infection (Fig. 5i–k). Notably, although flagellar stator units are widely distributed among both Gram-positive and Gram-negative bacteria¹³, Zorya is under-represented in bacteria with single-membrane cell envelopes, suggesting that differences in cell wall architecture or phage infection mechanisms might prevent effective phage-induced activation of Zorya (Extended Data Fig. 1a–c).

Sensing perturbation of the cell envelope provides an elegant mechanism that exploits the critical early stages of infection to trigger a direct anti-viral defence. While the exact mechanism of phage-induced ZorAB anchoring and activation remains unclear, the need of a rotary motor and the long ZorA tail to recruit and activate effectors suggests an intriguing hypothesis in which the effectors (ZorC/ZorD) are specifically activated in close proximity to the cell membrane. This localized effector function would protect host DNA from effector activity without relying on epigenetic-based self versus non-self discrimination mechanisms³¹ (Supplementary Discussion).

Overall, we provide structural and functional insights into the Zorya defence system and propose that Zorya acts early in infection by sensing perturbation of the cell envelope to initiate a localized anti-phage response near the cell membrane. Our work paves the way for further research to understand the detailed mechanisms of this unique activation signal for anti-phage defence.

Methods

Phylogenetic analysis of Zorya systems

To create the phylogenetic tree shown in Extended Data Fig. 1b, Zorya systems were identified by running PADLOC³⁴ v.1.1.0 with PADLOC-DB v.1.4.0 across RefSeq v209 bacterial genomes³⁵. For a representative sample of high-quality Zorya systems, the results were filtered for systems with canonical gene order (zorABCD, zorABE or zorFABG for types I, II and III, respectively) that were not at the edge of a contig and where the ZorA protein had minimum PADLOC-DB target and hmm alignment coverages of 80%. The respective ZorA sequences were trimmed to 250 amino acids to roughly isolate the head domain (representing the TMD motor unit). The ZorA head sequences were then clustered using MMseqs2³⁶ v.15.6f452 with the options --min-seq-id 0.8 --coverage 0.8, and the representative sequences for each resulting cluster were used for subsequent analyses. For the representative ZorA sequences, a diversified ensemble of 100 replicate alignments was built using the Super5 algorithm of muscle (v.5.1)³⁷ by perturbing each guide tree with 25 different seeds. E. coli MotA (WP_000906340.1) was included in the alignments as an outgroup. The alignment with the greatest column confidence was used to build a phylogenetic tree with FastTree v.2.1.11 with the option -lg (ref. ³⁸). Bacteria encoding each protein were assigned taxonomy based on GTDB v214.1³⁹, grouping phyla with alphabetic suffixes into their base phylum. To create the phylogenetic tree shown in Extended Data Fig. 1c, the GTDB v214.1 bacterial reference tree was filtered for genomes present in RefSeq v209 and collapsed to the phylum level.

Cloning of the Zorya defence system and mutagenesis

The EcZorI operon with its native promoter region was amplified by PCR from the E. coli strain NCTC9026 genome (purchased from the National Collection of Type Cultures (NCTC)) and subcloned into a modified pACYC vector using the In-Fusion cloning strategy (In-Fusion Snap Assembly Master Mix; TaKaRa, 638947). The PaZorI operon was amplified from the P. aeruginosa strain DSM24068 genome (DSMZ-German Collection of Microorganisms and Cell Cultures; Leibniz Institute) and was subcloned into a modified pACYC vector under the E. coli ZorI native promotor using the In-Fusion cloning strategy. For generating mutations (point mutations, deletions, mNeonGreen or Halo tag insertions, where EcZorI zorC or zorCD genes were replaced by PaZorI zorC or zorCD), plasmids were constructed based on standard cloning techniques (In-fusion snap assembly). All plasmids were verified by either Sanger or Nanopore sequencing.

Phage infectivity assays

The host E. coli ΔRM (hereafter ΔRM, a derivative of E. coli MG1655 engineered to remove multiple restriction modification systems, which was used isolate the BASEL phage collection)¹⁵, possessing either pControl (pACYC) or pEcZorI (or mutants thereof) were grown overnight in LB + chloramphenicol (25 µg ml⁻¹). EOP assays were performed using bacterial lawns of the host strain in 0.35% LB agar + 10 mM MgSO₄ + 2 mM CaCl₂ overlaid onto 1.5% LB agar + chloramphenicol. Tenfold dilution series of phages were spotted onto the overlays, air-dried, then the plates were incubated overnight at 30 °C. Liquid culture infection time courses were performed in 96-well plates in an incubated shaking plate reader at 30 °C. The time courses were begun with cells at an optical density at 600 nm (OD₆₀₀) of 0.05 and phages were added at the indicated MOI, assuming an OD₆₀₀ to cell ratio of 3 × 10⁸ cells per OD₆₀₀ unit. The average amino acid identity between phages (indicative of relatedness) and hierarchical clustering of phage genomes was calculated using EzAAI⁴⁰, separating clusters with proteome coverage <40%.

Phage adsorption and one-step growth curve assays

Overnight cultures of ΔRM possessing either pControl or pEcZorI were used to inoculate fresh LB + chloramphenicol cultures at a 1:100 dilution. The inoculated cultures were grown at 30 °C with shaking until reaching an OD₆₀₀ of 0.4–0.6, then collected by centrifugation, washed with LB + chloramphenicol and resuspended at an OD₆₀₀ of 1.0 in LB + 10 mM MgSO₄ + 2 mM CaCl₂. For phage adsorption assays, 10 ml samples of resuspended cells were infected with phage Bas24 at an MOI of 10⁻⁴, then the samples were mixed and incubated at 30 °C without shaking. For the 0 min timepoint (total input phages), 100 µl samples were removed and added to 0.35% LB agar seeded with ΔRM + pControl (as an indicator lawn), then poured on top of 1.5% LB agar + chloramphenicol. For each subsequent timepoint, 1 ml samples were taken, centrifuged to pellet cells and the supernatant (containing unabsorbed phages) was then filtered through a 0.2 µm PES syringe filter. Samples (100 µl) of the filtered supernatant were added to indicator overlays (as above) poured onto 1.5% LB agar + chloramphenicol. All overlay plates were incubated overnight at 30 °C before counting plaques. For each timepoint, the percentage of unabsorbed phages was calculated as the timepoint plaque count/plaque count for the time 0 min pControl sample. For the one-step phage growth curves (burst time and size), 2 ml samples of the cells resuspended at and OD₆₀₀ of 1.0 in LB + 10 mM MgSO₄ + 2 mM CaCl₂ (as above) were infected with phage Bas24 at an MOI of 10⁻⁴, then the samples were mixed and two tenfold diluted samples were prepared, and the dilution series for each sample was then incubated at 30 °C without shaking. At the indicated timepoints, 100 µl samples of each dilution were removed and added to 0.35% LB agar seeded with ΔRM + pControl (as an indicator lawn), then poured on top of 1.5% LB agar + chloramphenicol. All overlay plates were incubated overnight at 30 °C before counting plaques. For each timepoint, the PFU was normalized to the PFU of the 0 min pControl samples.

Conjugation assays

Plasmids encoding kanamycin resistance and each possessing different origins of replications (ColE1, pMAT16; RSF1010, pPF1825⁴¹; pBBR1, pSEVA237R⁴²; or RK2, pPF1619) were conjugated from the E. coli donor ST18^43,44 (an auxotroph requiring supplementation with 5-aminolevulinic acid; ALA) into the E. coli recipient ΔRM possessing either pControl or pEcZorI. Matings were performed at the indicated donor to recipient ratios (D:R) and incubated overnight on LB agar + chloramphenicol + ALA at 30 °C. The conjugation efficiency was determined by plating dilution series of the matings onto LB agar + chloramphenicol + kanamycin (transconjugants) and LB agar + chloramphenicol (total recipients). The transconjugant frequency was defined as the transconjugant CFU/recipient CFU.

Transformation assays

Chemically competent cells of ΔRM possessing either pControl or pEcZorI were prepared according to the Inoue method⁴⁵, with HEPES-KOH pH 6.8 used for the transformation buffer. Cells were stored in 200 µl aliquots at −80 °C before use. For each transformation assay, 5 ng of plasmid (quantified using a Qubit BR kit) was used. Plasmids used were as described above for the conjugation assays (ColE1, pMAT16; or pBBR1, pSEVA237R).

Cell survival assays

Overnight cultures of ΔRM possessing either pControl or pEcZorI were used to inoculate fresh LB + chloramphenicol cultures at a 1:100 dilution. The inoculated cultures were grown at 30 °C with shaking until reaching an OD₆₀₀ of 0.4–0.6, then collected by centrifugation, washed with LB + chloramphenicol, and resuspended at an OD₆₀₀ of 0.2. Phage Bas24 was then added at an MOI of 5 to each sample; control samples without phage addition were also included. After 20 min adsorption, tenfold serial dilutions of each sample were plated (100 µl each) onto LB + chloramphenicol, then incubated overnight at 30 °C. The cell survival rate was calculated as [CFU obtained + Bas24]/[CFU obtained without phage addition].

Phages and phage genome purification

Phage primary stocks were prepared using the double-agar method, with near-confluent plaque overlays. The phages were collected by adding SM buffer (100 mM NaCl, 8 mM MgSO₄, 50 mM Tris-HCl pH 7.5, 5 mM CaCl₂) on top of the overlay agar and mixed for 4 h at 4 °C. The suspension was collected and centrifuged for 15 min at 4,000g. High-titre phage samples were obtained by inoculating 1–3 l of LB with a 10³ dilution of an overnight culture of ∆RM and grown at 37 °C to an OD₆₀₀ of 0.3. The bacterial culture was inoculated with the primary stock to an MOI of 0.025 and infection was carried out at 37 °C at 90 rpm until a clear lysate was obtained. The lysate was collected at 4,000g, for 15 min at 4 °C. After decanting the supernatant, 1 µg ml⁻¹ of DNase I and 1 µg ml⁻¹ of boiled RNase A were added to the cleared lysate. The lysate was gently stirred at 90 rpm for 30 min at room temperature.

Phages were concentrated by polyethylene glycol (PEG) precipitation. NaCl was gradually added to a final concentration of 1 M, followed by gradual addition of 10% PEG 8,000 with continuous stirring at room temperature until dissolved. After obtaining a clear solution, the lysate was stirred (100 rpm, 30 min; 4 °C) and left overnight at 4 °C. The lysate was centrifuged (15,000g, 1 h, 4 °C) and the clear supernatant was removed. The precipitate was resuspended in the minimal amount (up to 2 ml) of SM buffer that allowed solubilization. Insoluble materials were removed by adding 20% (v/v) of chloroform and centrifuged (8,000g, 10 min). The supernatant was stored at 4 °C to be used as phage sample for the next step. The phage was then purified by rate zonal separation using OptiPrep Density Gradient Medium (Sigma-Aldrich) in a density gradient ranging from 50 to 10%, diluted in SM medium. The phage sample was applied on the top of the gradient and centrifuged (150,000g, 18 h, 4 °C). The phage was extracted, dialysed against SM buffer and the samples were stored at 4 °C. The phage genomes were extracted using the Phage DNA isolation kit from Norgen Biotek, aliquoted and stored at −20 °C.

Phage genome labelling

Stocks of Bas24 were treated with Pierce Universal Nuclease according to the manufacturer’s protocol for 1 h at 37 °C. To stain the phage genomic DNA, SYTOX Orange (Invitrogen) stock solution was added to 10 ml of the phage lysate at a concentration of 1:2,000 and incubated overnight at 4 °C in the dark. Stained phage particles were subsequently purified by PEG precipitation. PEG 6,000 was added to the lysate to a final concentration of 10% (w/v) and incubated overnight at 4 °C to allow for phage aggregation and precipitation. The lysate was centrifuged at 4,000g for 30 min at 4 °C to pellet the phages and the supernatant was carefully discarded without disturbing the phage pellet. The phage pellet was then washed by gently adding a 1 ml SM buffer, centrifuged at 6,000g for 2 min and used for subsequent time-lapse microscopy experiments.

Protein expression and purification

ZorAB

The full-length genes of E. coli ZorA and ZorB code for 729 and 246 residues, respectively. The tandem gene was PCR amplified from the E. coli strain NCTC9026 genome and subcloned into a modified pET vector containing a C-terminal human rhinovirus (HRV) 3C protease cleavage site and a twin-Strep-tag II (resulting in pET11a-ZorA-ZorB-3C-TSII). The plasmids containing the recombinant genes were transfected into E. coli C43(DE3) competent cells and the proteins were expressed in LB medium. When the culture OD₆₀₀ reached 0.6–0.8, the temperature was decreased from 37 °C to 24 °C, then grown until the OD₆₀₀ reached approximately 0.8–1.0, before 0.5 mM isopropyl β-d-1-thiogalactopyranoside (IPTG) was added for overnight protein induction. The culture was collected, and the cell pellet was resuspended in buffer A containing 20 mM HEPES-NaOH pH 7.5, 300 mM NaCl supplemented with EDTA-free protease inhibitor (Thermo Fisher Scientific) and lysozyme from chicken white egg (Sigma-Aldrich) to a final concentration of 50 μg ml⁻¹ and deoxyribonuclease I from bovine (Sigma-Aldrich) to a final concentration of 30 μg ml⁻¹. The mixture was disrupted by high-pressure homogenizer and centrifuged at 185,000g for 1 h. The pellet containing the membrane was collected and solubilized using buffer B containing 30 mM HEPES-NaOH pH 7.5, 300 mM NaCl, 10% glycerol, 2% LMNG (Anatrace), supplemented with EDTA-free protease inhibitor at 4 °C for 2 h. The solubilized membrane was then centrifuged at 90,000g for 40 min and the supernatant was loaded onto a gravity flow column containing 2 ml (resin volume) of Strep-Tactin Superflow high-capacity resin (IBA), pre-equilibrated with wash buffer containing 20 mM HEPES-NaOH pH 7.5, 300 mM NaCl, 10% glycerol and 0.005% LMNG. The resins were washed five times with 2–3 resin volumes of the wash buffer and elution was carried out five times with 0.5 resin volume (1 ml) of elution buffer containing 20 mM HEPES-NaOH pH 7.5, 300 mM NaCl, 10% glycerol, 0.005% LMNG and 10 mM desthiobiotin). The recombinant protein was then concentrated and loaded onto a pre-equilibrated (20 mM HEPES-NaOH pH 7.5, 150 mM NaCl, 0.002% LMNG) Superose 6 Increase 10/300 GL size-exclusion chromatography column. The fractions from the elution peak corresponding to the molecular mass of the ZorAB complex were pooled, and the protein was concentrated for cryo-EM grid preparation and functional experiments. The procedures for expression and purification of ZorAB mutants were similar to those for the ZorAB wild type.

ZorB and MotB soluble PGBDs

The PGBDs of each ZorB and ZorB(Y151A/N152A/L155A/R159A) (from Arg52 to C terminus) and E. coli MotB (from Glu96 to C terminus) were cloned from the existing full-length expression constructs¹¹ into pET11a by PCR amplification-based deletion mutagenesis (TakaraBio). The ZorB and MotB PGBDs were purified similarly to ZorC and ZorD with a few modifications. The ZorB and the ZorB(Y151A/N152A/L155A/R159A) PGBD vectors were transformed into Rosetta-gami-2(DE3) competent E. coli (Novagen). Cells were grown in LB medium supplemented with 100 μg ml⁻¹ ampicillin, 34 μg ml⁻¹ chloramphenicol and 10 μg ml⁻¹ tetracycline at 37 °C to an OD₆₀₀ of 0.7. The cells were then induced with 0.5 mM IPTG and allowed to grow for 16 h at 18 °C. The cultures were collected and the cell pellets resuspended in lysis/wash buffer (50 mM Tris-HCl pH 8.0, 150 mM NaCl supplemented with EDTA-free protease inhibitor cocktail (Roche)). Moreover, 1 mg of DNase I and 0.5 mM MgCl₂ was added to the resuspended cells. The cells were lysed using the Avestin Emulsiflex C3 homogenizer, cooled to 4 °C and soluble lysates were cleared by centrifugation at 30,000g at 4 °C for 30 min. The supernatant was then run over a gravity-flow column containing 2 ml (resin volume) of Strep-Tactin 4Flow high-capacity resin (IBA), pre-equilibrated with wash buffer (50 mM Tris-HCl pH 8.0, 150 mM NaCl). The resin was washed with 20 ml lysis/wash buffer and protein was eluted in 12 ml elution buffer (50 mM Tris-HCl pH 8.0, 150 mM NaCl, 10 mM desthiobiotin). The elution was then concentrated and run over a Superose 6 Increase 10/300 GL size-exclusion chromatography column into gel-filtration buffer (20 mM Tris-HCl pH 8.0, 150 mM NaCl). The MotB PGBD was expressed and purified identically to the ZorB PGBDs, with the exception that it was expressed in BL21(DE3) gold E. coli in LB medium supplemented with 100 μg ml⁻¹ ampicillin.

ZorC

The predicted zorC gene encodes 560 residues. The zorC gene together with a short region upstream of the zorC N terminus that encodes 7 residues (LPVGYAT) was PCR amplified from the DNA genome of E. coli strain NCTC9026 and subcloned into the modified pET vector (resulting in pET11a-ZorC-3C-TSII). E. coli BL21 (DE3) gold chemically competent cells were transformed with the plasmids and the protein was expressed in LB medium with the presence of 100 μg ml⁻¹ of ampicillin. In brief, when the OD₆₀₀ reached 1.0–1.2, the temperature was decreased to 16 °C and 0.5 mM IPTG was added for overnight protein induction. The culture was collected, and the cell pellet was resuspended using buffer containing 20 mM Tris-HCl pH 7.5, 10% glycerol and 500 mM NaCl supplemented with EDTA-free protease inhibitor (Thermo Fisher Scientific). The cells were lysed using an Avestin Emulsiflex C3 homogenizer, cooled to 4 °C and centrifuged at 18,000g for 40 min. The supernatant was then added to a gravity-flow column containing 3 ml (resin volume) of Strep-Tactin Superflow high-capacity resins (IBA), pre-equilibrated with wash buffer (20 mM Tris-HCl pH 7.5, 10% glycerol and 500 mM NaCl). Resins were washed five times with 2–3 resin volumes of wash buffer and elution was performed with 4 CV of elution buffer (20 mM Tris-HCl pH 7.5, 500 mM NaCl and 10 mM desthiobiotin). The recombinant protein was pooled and concentrated and was loaded onto a pre-equilibrated (20 mM Tris-HCl pH 7.5, 500 mM NaCl) Superose 6 Increase 10/300 GL size-exclusion chromatography column. Peak fractions were pooled, and another round of size-exclusion chromatography was carried out with buffer 20 mM HEPES-NaOH pH 7.5, and 150 mM NaCl to decrease the NaCl concentration. The fractions from the elution peak corresponding to the molecular mass of ZorC were pooled and the protein was concentrated to approximately 1 mg ml⁻¹ for cryo-EM grid preparation and functional experiments. ZorC proteins used for EMSAs were exchanged into buffer containing 20 mM Tris-HCl pH 7.5, 300 mM NaCl, 10% glycerol and 1 mM TCEP after elution. Pure fractions were concentrated and flash-frozen in small aliquots and stored at −80 °C until use. The sample purity was assessed using SDS–PAGE. The procedures of expression and purification of ZorC mutants were similar to those for the ZorC wild type.

ZorD

The predicted zorD gene encoding 1,086 residues was PCR amplified from the DNA genome of E. coli strain NCTC9026 and was subcloned into the modified pET vector, resulting in pET11a-ZorC-3C-TSII. The expression and purification of ZorD protein were similar to those for the ZorC protein, except for the purification buffer. In brief, the suspension buffer contained 150 mM NaCl, 20 mM HEPES-NaOH pH 7.5 and 10% glycerol; the wash buffer was the same as the suspension buffer and the elution buffer contained 150 mM NaCl, 20 mM HEPES-NaOH pH 7.5, 10% glycerol and 10 mM desthiobiotin; and the size-exclusion chromatography buffer contained 150 mM NaCl and 20 mM HEPES-NaOH pH 7.5. Purified ZorD was concentrated to 0.4–0.6 mg ml⁻¹ for functional experiments and cryo-EM grid preparation. For the nuclease experiment, ZorD protein was kept in the elution buffer and flash-frozen in small aliquots and stored at −80 °C until use.

Cryo-EM grid preparation, data collection, model building and refinement

ZorAB

Freshly purified ZorAB sample was concentrated to 2–3 mg ml⁻¹ and 2.7 μl protein was applied onto glow-discharged (30 s, 5 mA) grids (Quantifoil R0.6/1 300 mesh Au) and plunge-frozen into liquid ethane using the Vitrobot Mark IV (FEI, Thermo Fisher Scientific) with the following settings: 100% humidity, 4°C, blotting force 25, 4–6 s blot time and 7 s wait time. Videos were collected using the semi-automated acquisition program EPU (FEI, Thermo Fisher Scientific) on the Titan Krios G2 microscope operated at 300 keV paired with a Falcon 3EC direct electron detector (FEI, Thermo Fisher Scientific). Images were recorded in electron-counting mode, at ×96,000 magnification with a calibrated pixel size of 0.832 Å and an underfocus range of 0.7 to 2.5 μm. The number of micrographs and the total exposure values for the different datasets are summarized in Supplementary Table 1. Grid preparation and data collection strategies for the ZorAB mutants were similar to those for the ZorAB wild type.

ZorC

Purified ZorC (3 μl at around 1 mg ml⁻¹) was applied onto glow-discharged (30 s, 5 mA) grids (UltrAuFoil R 0.6/1, 300 mesh, Gold) and plunge-frozen into liquid ethane using the Vitrobot Mark IV (FEI, Thermo Fisher Scientific), with the following settings: 100% humidity, 4 °C, blotting force 20, 4 s blot time and 10 s wait time. Videos were collected using the semi-automated acquisition program EPU (FEI, Thermo Fisher Scientific) on the Titan Krios G2 microscope operated at 300 keV paired with a Falcon 3EC direct electron detector (FEI, Thermo Fisher Scientific). Images were recorded in electron-counting mode, at ×96,000 magnification with a calibrated pixel size of 0.832 Å and an underfocus range of 1 to 2.5 μm. The number of micrographs and total exposure values for the datasets are summarized in Extended Data Table 1. For the ZorC + DNA sample, ZorC (final concentration 0.6 mg ml⁻¹) was mixed with commercial pUC19 plasmid (NEB) (final concentration of 0.5 μg μl⁻¹). The samples were incubated at room temperature for 30 min, followed by 30 min at 4 °C before grid preparation. The samples (3 μl) were applied to UltraAuFoil R 2/2, 200 mesh Gold grids (glow discharged 60 s at 10 mA) and plunge-frozen as described above, but with the following settings: blotting force 15, 3 s blot time.

ZorD

ZorD showed preferred orientation of particles on ice. Zwitterionic detergent (0.5% CHAPSO; Anatrace) was added to the purified sample to a final concentration of 0.0125% before cryo-EM grid preparation. For the apo form, the preparation of grids was similar to ZorC. For ZorD in complex with ATP-γ-S, 4 µl of 0.1 mM ATP-γ-S was added to 400 µl of purified ZorD at 0.0375 mg ml⁻¹. The mixture was concentrated to 15 µl to reach a ZorD concentration of around 0.6 mg ml⁻¹. The grid preparation was similar to for ZorC. The number of micrographs and total exposure values for the different datasets are summarized in Extended Data Table 1.

Cryo-EM data processing

All datasets were processed using cryoSPARC⁴⁶ v.4.2.1, unless otherwise stated. We started by using Patch motion correction to estimate and correct for full-frame motion and sample deformation (local motion). Patch contrast function (CTF) estimation was used to fit local CTF to micrographs. Micrographs were manually curated to remove low-quality data (relative ice thickness value greater than 1.1 and CTF value worse than 3.5 Å). We performed particle picking by template picking or using topaz particle picking⁴⁷. Particles were extracted with a box size of 500 pixels for ZorAB datasets, 256 pixels for the ZorC dataset and 400 pixels for the ZorD dataset. One round of 2D classification was performed followed by ab initio reconstruction. Heterogeneous refinement was used to exclude broken particles. Non-uniform refinement was applied with a dynamic mask to obtain a high-resolution map. Local refinement was additionally performed with a soft mask to achieve a higher-resolution map of some flexible regions. For all datasets, the number of videos, the number of particles used for the final refinement, map resolution and other values during data processing are summarized in the Extended Data Table 1.

Model building and validation

We used AlphaFold2 or AlphaFold3²⁴ to predict all of the initial models. The predicted models were manually fit into the cryo-EM density by using UCSF ChimeraX⁴⁸. The model was refined in Coot⁴⁹, or using StarMap⁵⁰ in the case of ZorC, for which the map is anisotropic and the resolution is modest. The model was then refined against the map using PHENIX real space refinement⁵¹. The ZorAB composite model was constructed by extending the pentameric tail as an idealized right-handed super-helical coiled-coil. Local conformations were manually adjusted in PyMol⁵² (v.2.5) and optimized through energy minimization using GROMACS (v.2022.5)⁵³ with CHARMM27 force field. However, it is worth noting an irregularity in the AF2 model, specifically in residues 312 to 322, which introduces a substantial twist in the ZorA tail, raising possibilities of other pentameric forms of the ZorA tail and further reflecting its conformational dynamics.

PG purification and pull-down experiments

PG was purified from E. coli ∆RM with the protocol adapted from a previous study⁵⁴_. In brief, ∆RM cells were incubated in 1 l LB medium until the OD₆₀₀ reached 0.8. Cells were collected and resuspended in 12 ml PBS buffer and split into two 50 ml Falcon tubes, then 10% (w/v) SDS solution (in PBS) was added to a final concentration of 6% (w/v). The Falcon tubes were boiled for 1 h with stirring at 500 rpm. The heat was turned off and the tube was allowed to cool to ambient temperature overnight. The next day, the solutions from both Falcon tubes were pooled into one 50 ml Falcon tube, and centrifuged at room temperature for 45 min at 108,000g. The pellet was washed five times with 5 ml Milli-Q water. The PG was resuspended in 20 ml of buffer containing 50 mM Tris-HCl pH 7.0, and α-amylase was added (Sigma-Aldrich) to a final concentration of 100 μg ml⁻¹ and incubated for 2 h at 37 °C. Next, 50 μg ml⁻¹ RNase A (Roche) and 10 μg ml⁻¹ DNase (Sigma-Aldrich) were added and incubated for an additional 2 h at 37 °C. The mixture was then supplemented with 20 mM MgSO₄, 10 mM CaCl₂ and 100 μg ml⁻¹ trypsin (Sigma-Aldrich), and incubated at 37 °C overnight. The next day, EDTA at pH 8 was added to a final concentration of 10 mM and 10% (w/v) SDS solution to a final concentration of 1% (w/v). The mixture was boiled for 20 min in a water bath and allowed to cool to ambient temperature. The tube was centrifuged at 108,000g for 45 min. The resulting pellet was washed five times with Milli-Q water to remove residual SDS. Finally, the pellet was resuspended in 300 μl of Milli-Q water, aliquoted into 35 μl portions and stored at −20 °C. For PG pull-down assays, the purified PG was washed with 1 ml PBS + 0.002% LMNG buffer and centrifuged at 20,000g for 30 min. Purified ZorAB and ZorAB ZorB(Y151A/N152A/L155A/R159A) mutant (10 μl, at a concentration of 2 mg ml⁻¹; ZorAB ZorB(Y151A/N152A/L155A/R159A) mutant is less stable, requiring the use of freshly purified protein) was incubated with the PG at room temperature for 1 h, then centrifuged at 20,000g for 30 min. The pellet was washed three times with 700 µl of the pull-down buffer by mixing and centrifugation (10 min, 12 °C, 20,000g). The supernatant was retained for SDS gel analysis. The pellet was resuspended with 20 μl of buffer and 5 μl of loading dye was added for SDS gel analysis.

For the pull-downs of ZorB PGBD, mutant ZorB PGBD (ZorB(Y151A/N152A/L155A/R159A) PGBD), MotB PGBD (positive control) and ZorE (negative control), the protocol was similar to as above. In brief, PG (PGN-ECndi ultrapure peptidoglycan (InvivoGen); due to the low yield of the laboratory-purified PG) was washed and resuspended in resuspension buffer (20 mM potassium phosphate, pH 7, 150 mM NaCl). Each pull-down reaction contained 10 µl of washed 25 mg ml⁻¹ PG, 4 µl of the indicated protein (each added from a 5 mg ml⁻¹ stock) and the pull-down buffer (20 mM potassium phosphate pH 6, 150 mM NaCl, 2 mM MgCl₂, 2 mM CaCl₂) to final volume of 100 µl. The samples were incubated for 30 min at 20 °C with gentle mixing. The insoluble PG was pelleted by centrifugation at 20,000g at 12 °C for 10 min, and the soluble supernatant was retained for SDS–PAGE analysis. The pellet was washed three times with 700 µl of the pull-down buffer by mixing and centrifugation (10 min, 12 °C, 20,000g). Finally, the pellet was resuspended in 100 µl of pull-down buffer and 15 µl of each sample (soluble supernatant and resuspended pellet) was mixed with 3 µl loading dye for SDS–PAGE analysis.

DNA-binding experiments

Frozen aliquots of ZorC and ZorC mutants in ZorC buffer (20 mM Tris-HCL pH 7.5, 300 mM NaCl, 10% glycerol, 1 mM TCEP) were thawed and centrifuged to remove possible aggregates. ZorC was diluted to 10× stocks for each concentration used, in the same ZorC buffer. The final EMSA reaction buffer contained phosphate-buffered saline (15.2 mM NaHPO₄, 0.90 mM CaCl₂, 2.7 mM KCl, 1.47 mM KH₂PO₄, 8.1 mM Na₂HPO₄, 0.49 mM MgCl₂, 137.9 mM NaCl) at pH 7.4 and 10% glycerol. ZorC or ZorC buffer was then added, followed by DNA substrate (10× stock in H₂O) or H₂O. DNA oligos with 5′-FAM modifications were synthesized by TAG Copenhagen and double-stranded constructs were obtained by annealing the unlabelled complement sequence (the dsDNA sequences used in this study are shown in Extended Data Fig. 7h). All components were incubated at 4 °C for 30 min and loaded onto a 1.5% (w/v) agarose gel made with 20 mM sodium phosphate buffer (pH 7.2). The samples were run for 30 min at 100 V and 4 °C using 20 mM sodium phosphate (pH 7.2) as the running buffer. The gels were visualized using the Odyssey XF Imaging System at 600 nm.

Nuclease assays

ZorD was incubated with 200 ng pUC19 (linearized by KpnI) in the reaction buffer containing 1× Cutsmart buffer (NEB) and 2 mM ATP (NEB) in a total volume of 25 μl. The reactions were incubated at 37 °C for 1 h with shaking at 600 rpm using the Eppendorf ThermoMixer. DNA product was purified using the NucleoSpin Gel and PCR Clean-up kit (Machery Nagel) using the standard protocol and was analysed with 1% E-Gel EX. For the reaction with the phage genomes, 200 nM proteins were incubated with around 100 ng purified phage genomic DNA in the same reaction buffer indicated above. The reactions were terminated by adding 1× E-gel loading buffer and product was analysed with 1% E-Gel EX.

MS sample preparation

Overnight cultures of E. coli ∆RM transformed with pEcZorI (or pControl), were used to inoculate (at a 1:1,000 dilution) 3 ml LB medium with antibiotics, then grown to an OD₆₀₀ of approximately 0.4. The cell pellet was collected, resuspended in 500 μl of 0.2 M Tris-HCl pH 8.0, and incubated for 20 min. Next, 250 μl of buffer (0.2 M Tris-HCl pH 8.0, 1 M sucrose and 1 mM EDTA) was added, along with 3 μl of 10 mg ml⁻¹ lysozyme. The mixture was incubated for 30 min, and 250 μl of 6% (w/v) SDS was added to a final concentration of 1%, after which the sample was heated at 99 °C for 10 min. The mixture was sonicated (Misonix Ultrasonic Liquid Processor with microtip Probe) to fragment DNA and RNA with the following settings: amplitude 10, time 5 s sonication and 5 s pause, 5 cycles.

For MS analysis, we performed protein aggregate capture digestion of proteins⁵⁵. To this end, 250 μl of bacterial lysate was taken from the total sample, and 750 μl of acetonitrile was added into the mixture, along with 50 µl magnetic microspheres that had been prewashed with PBS buffer. The mixture was allowed to settle for 10 min, before retention of the magnetic microspheres by a magnetic plate. Beads were washed once with 1 ml acetonitrile and once with 1 ml of 70% ethanol, after which all ethanol was removed and the beads were stored at −20 °C until further processing. The frozen beads were thawed on ice, supplemented with 100 µl ice-cold 50 mM Tris-HCl pH 8.5 buffer supplemented with 2.5 ng μl⁻¹ Lys-C and gently mixed (on ice) every 5 min for 30 min. Digestion was performed for 3 h using the Eppendorf ThermoMixer with shaking at 1,250 rpm and 37 °C. Next, the beads were chilled on ice and 250 ng of sequencing-grade trypsin was added, after which samples were gently mixed (on ice) every 5 min for 30 min. Final digestion was performed overnight using the Eppendorf ThermoMixer with shaking at 1,250 rpm and 37 °C. Peptides were separated from magnetic microspheres using 0.45 µm filter spin columns, and peptides were reduced and alkylated by adding TCEP and chloroacetamide to 5 mM for 30 min before peptide clean-up using the low-pH C18 StageTip procedure. C18 StageTips were prepared in-house by layering four plugs of C18 material (Sigma-Aldrich, Empore SPE Disks, C18, 47 mm) per StageTip. Activation of StageTips was performed with 100 μl 100% methanol, followed by equilibration using 100 μl 80% acetonitrile in 0.1% formic acid, and two washes with 100 μl 0.1% formic acid. Samples were acidified to pH < 3 by addition of trifluoroacetic acid to a concentration of 1%, after which they were loaded on StageTips. Subsequently, StageTips were washed twice using 100 μl 0.1% formic acid, after which peptides were eluted using 80 µl 30% acetonitrile in 0.1% formic acid. All fractions were dried to completion using a SpeedVac at 60 °C. Dried peptides were dissolved in 25 μl 0.1% formic acid and stored at −20 °C until MS analysis.

Approximately 1 µg of peptide was analysed per injection. All samples were analysed on the EASY-nLC 1200 system (Thermo Fisher Scientific) coupled to the Orbitrap Astral mass spectrometer (Thermo Fisher Scientific). The samples were analysed on 20-cm-long analytical columns, with an internal diameter of 75 μm, and packed in-house using ReproSil-Pur 120 C18-AQ 1.9 µm beads (Dr Maisch). The analytical column was heated to 40 °C, and elution of peptides from the column was achieved by application of gradients with stationary phase buffer A (0.1% formic acid) and increasing amounts of mobile phase buffer B (80% acetonitrile in 0.1% formic acid). The primary analytical gradient ranged from 10% B to 38% B over 57.5 min, followed by a further increase to 48% B over 5 min to elute any remaining peptides, and by a washing block of 15 min. Ionization was achieved using a NanoSpray Flex NG ion source (Thermo Fisher Scientific), with a spray voltage of 2 kV, ion transfer tube temperature of 275 °C and RF funnel level of 50%. All full precursor (MS1) scans were acquired using the Orbitrap mass analyser, while all tandem fragment (MS2) scans were acquired in parallel using the Astral mass analyser. Full scan range was set to 300–1,300 m/z, MS1 resolution to 120,000, MS1 AGC target to 250 (2,500,000 charges) and MS1 maximum injection time to 150. Precursors were analysed in data-dependent acquisition mode, with charges 2–6 selected for fragmentation using an isolation width of 1.3 m/z and fragmented using higher-energy collision disassociation with normalized collision energy of 25. Monoisotopic precursor selection was enabled in peptide mode. Repeated sequencing of precursors was minimized by setting expected peak width to 20 s, and dynamic exclusion duration to 20 s, with an exclusion mass tolerance of 10 ppm and exclusion of isotopes. MS2 scans were acquired using the Astral mass analyser. MS2 fragment scan range was set to 100–1,500 m/z, MS2 AGC target to 50 (5,000 charges), MS2 intensity threshold to 50,000 charges per second and MS2 maximum injection time to 5 ms, therefore requiring a minimum of 250 charges for attempted isolation and identification of each precursor. Duty cycle was fixed at 0.3 s, acquiring full MS scans at ~3.3 Hz and with auto-fitting of Astral scans resulting in MS2 acquisition at a rate of ~100–200 Hz.

MS data analysis

All RAW files were analysed using MaxQuant software (v.2.4.3.0)⁵⁶, the earliest release version to support Astral RAW files. The default MaxQuant settings were used, with exceptions outlined below. For generation of the in silico spectral library, the four full-length Zorya protein sequences were entered into a FASTA database, along with all (23,259) Swiss-Prot-reviewed E. coli sequences (taxonomy identifier 562) downloaded from UniProt⁵⁷ on 7 September 2023. The data were first searched using pFind (v.3.2.0)⁵⁸, using the Open Search feature to determine the overall peptide properties and commonly occurring (affecting >1% of PSMs) peptide modification in an unbiased manner. For searching Astral.RAW files using pFind, .RAW files were first converted to .mzML using OpenMS (v.3.0.0)⁵⁹. For the main data search using MaxQuant, digestion was performed using ‘Trypsin/P’ with up to 2 missed cleavages (default), with a minimum peptide length of 6 and a maximum peptide mass of 5,000 Da. No variable modifications were considered for the first MS/MS search, which is only used for precursor mass recalibration. For the MS/MS main search a maximum allowance of three variable modifications per peptide was set, including protein N-terminal acetylation (default), oxidation of methionine (default), deamidation of asparagine, peptide N-terminal glutamine to pyroglutamate and replacement of three protons by iron (cation Fe[iii]) on aspartate and glutamate. Unmodified and modified peptides were stringently filtered by setting a minimum score of 10 and 20, and a minimum delta score of 20 and 40, respectively. First-search mass tolerance was set to 10 ppm, and the maximum charge state of considered precursors was set to 6. Label-free quantification (LFQ) was enabled, Fast LFQ was disabled. iBAQ was enabled. Matching between runs was enabled with a match time window of 1 min and an alignment time window of 20 min. Matching was allowed only between same-condition replicates. Data were filtered by posterior error probability to achieve a false-discovery rate of <1% (default), at the peptide-spectrum match, protein assignment and site-decoy levels.

MS data statistics

All statistical data handling was performed using the Perseus software⁶⁰, including data filtering, log₂-transformation, imputation of missing values (down shift 1.8 and width 0.15) and two-tailed two-sample Student’s t-testing with permutation-based false-discovery rate control. To determine the relative concentration of all proteins in the samples, LFQ-normalized intensity values for each protein were adjusted by molecular mass. To approximate the absolute copy numbers, we extracted known protein copy numbers based on the ‘LB’ condition as reported previously⁶¹, log₂-transformed them and aligned them to the molecular-mass-adjusted LFQ intensity values from our own data, resulting in 1,901 out of 2,418 quantified protein groups receiving a known copy-number value (R² = 0.6129). We next subtracted the overall median from all log₂ values and determined the absolute delta between the values of each pair. Out of all pairs, 459 had a log₂ delta of <0.5, which we considered to be a ‘proteomic ruler’. Linear regression was performed on the remaining pairs (R² = 0.9868) to determine a conversion factor between molecular-mass-adjusted LFQ intensity and absolute copy number.

TIRF microscopy cultivation conditions

Overnight cultures of E. coli strains expressing ZorB–HaloTag and ZorC or ZorD translational fusions to mNG were incubated shaking at 180 rpm in LB Lennox containing 20 mM MgSO₄, 5 mM CaCl₂ and supplemented with 12.5 µg ml⁻¹ chloramphenicol at 30 °C. The next day, a subculture was inoculated 1:100 and grown at 30 °C until an OD₆₀₀ between 0.3–0.5 was reached. Subsequently, cells were diluted to an OD₆₀₀ of 0.2. For HaloTag fusions, cells were washed once in PBS supplemented with a final concentration of 0.2% glucose and stained with 1 µM TMR ligand for 30 min. To remove excess of TMR after staining, cells were washed twice with PBS supplemented with 0.2% glucose. Cells were then exposed to phages at indicated MOIs or incubated untreated for 30 min in a 2 ml Eppendorf tube under shaking conditions (<650 rpm in an Eppendorf ThermoMixer) at 30 °C. For TIRF microscopy, 1 µl of cells and phage mix was spotted onto an agarose pad (1.2% in Milli-Q water of UltraPure agarose, Invitrogen) and directly imaged.

TIRF microscopy acquisition and data evaluation

TIRF microscopy was performed using the Nikon Eclipse Ti2 inverted microscope equipped with the ILAS 2 TIRF module (Gataca Systems) and a TIRF ×100/1.49 NA oil objective. The samples were excited at 50 ms exposure for 14 frames using a 488 nm laser (power of 550 µW with a sensor area of 283.5 mm²) at 80% and emission was recovered through a quad TIRF filter cube (emission: 502–549 nm). For the HaloTag constructs, five frames were acquired using a 561 nm laser (power of 930 µW with a sensor area of 283.5 mm²) at 80% with an exposure time of 50 ms. Emission was recovered through a quad TIRF filter cube (emission: 581–625 nm, followed by an emission wheel filter: 580–611 nm). The second frame (for HaloTag constructs, the third frame) in the fluorescent channel of the acquired TIRF microscopy images was denoised using the Nikon software package Denoise.ai. Subsequently, fluorescent maxima of ZorB, ZorC or ZorD translational fusions to mNG were detected using MicrobeJ⁶² run in Fiji⁶³. For the co-localization analysis of ZorB–HT with either ZorC–mNG or ZorD–mNG, ZorB focus detection was performed using ilastik⁶⁴. Subsequently, the obtained ZorB binary masks were overlayed in MicrobeJ onto the original image and cells were manually detected in MicrobeJ. From this, parameters such as ZorB focus count, intensity in both HaloTag and mNG channel per cell and mNG signal of the entire cell body were extracted. We defined co-localization of ZorB with a cytoplasmic Zorya component (ZorC or ZorD) if mNG signal within the detected ZorB complex area was 1.5-fold greater than the average cytoplasmic mNG fluorescence of the complete cell. Finally, cells containing at least one ZorB with either ZorD or ZorC were considered to be co-localized. Statistics were calculated in Prism GraphPad 9 using the in-built unpaired t-tests or one-way ANOVA.

Time-lapse microscopy and data evaluation

Cells were prepared as stated above for TIRF microscopy. In the case of the Bas54-parS phage infection experiments, the E. coli cells expressed an IPTG-inducible ParB^P1-mScarlet fusion protein from pALA2703-mScarlet-ParB^P1 in addition to EcZorI or the empty vector control⁶⁵. After diluting the subculture to an OD₆₀₀ of 0.2, cells were exposed to phages at an MOI of 5 (or unknown if labelled) for 5 min (or none) under shaking conditions and, subsequently, 1 µl was spotted onto a 1.2% (w/v) agarose pad (dissolved in LB:MQ with a ratio of 1:5). Expression of ParB^P1–mScarlet was induced by addition of 500 µM IPTG in the subculture. For the time-lapse experiments involving CCCP, cells were exposed to 30 µM CCCP and Bas24 (MOI of 5) for 30 min (or CCCP only) under shaking conditions and followed by spotting 1 µl onto an agarose pad (composition as above, supplemented with 30 µM CCCP). Microscopy slides were then mounted into an incubation chamber (preheated to 37 °C) surrounding the microscope. Acquisition was performed using the Nikon Eclipse Ti2 inverted microscope equipped with a CFI Plan Apochromat DM ×60 Lambda oil Ph3/1.40 objective and phase-contrast images were obtained every 2 min for the indicated time span. Obtained time-lapse images were segmented using ilastik to detect the cell areas⁶³. Time-lapse microscopy videos of DNA-labelled Bas24 and Bas54-parS phage infection experiments were x,y drift-corrected using the Fiji Fast4DReg plugin^66,67. The intensity decay due to photobleaching of the SYTOX Orange fluorescent channel was subsequently corrected using the Bleach Corrector Fiji plugin⁶⁸. The final graphs and videos were prepared using a custom Python script.

Construction of the phage Bas54-parS

A parS site was introduced between gp69 and gp70 of phage MaxBurger (Bas54)¹⁵ by homologous recombination with a synthetic template followed by CRISPR–Cas13a selection against the parental wild type using a setup similar to the procedure described elsewhere⁶⁹. In a first step, the parS sequence (TCGCCATTCAAATTTCACTATTAACTGACTGTTTTTAAAGTAAATTACTCTAAAATTTCAAGGTGAAATCGCCACGATTTCAC) was inserted between plasmid-encoded homology arms of 100–200 bp length by PCR⁷⁰, creating pAH210_Bas54_parS_H1. Subsequently, an E. coli K-12 host was transformed with this plasmid and then infected with phage Bas54 to enable homologous recombination. The lysate from this infection—containing wild-type and recombinant phages—was then subjected to CRISPR–Cas13a selection against the parental wild type using a crRNA targeting CTCTGAAGACCTCCAGTAGTAAGATGTAAGT (5′–3′), which includes the 3′ end of gp69 and the downstream region, which is disrupted by insertion of the parS site. CRISPR–Cas13a selection was performed using a two-plasmid setup of pAH221 (expressing LbuCas13a) and pAH218_LbuCas13a_parS_H1 (expressing the crRNA). Plaques growing after CRISPR–Cas13a selection were screened for successful insertion of the parS site by PCR and the sequence of the recombinant was confirmed by Sanger sequencing.

Bioinformatic analyses of ZorA motor and tail lengths

PADLOC (v.1.1.0)³⁴ with PADLOC-DB (v.1.4.0) was used to identify Zorya systems in RefSeq v209³⁵ bacterial and archaeal genomes. Of the systems identified, we excluded those containing pseudogenes or more than one copy of each Zorya gene, or systems with non-canonical gene arrangements (Zorya genes are typically on the same strand, in type-specific conserved gene orders, for example, zorABCD, zorABE or zorGABF for types I–III, respectively). To reduce redundancy due to highly related genome sequences in the RefSeq database, we next selected representative Zorya systems by first clustering the sequences (using MMseqs2 (v.14.7e284)³⁶ with the options --min-seq-id 0.3 --coverage 0.8) of the proteins encoded by the three adjacent open reading frames on either side of each Zorya system, then randomly selecting one representative system for each unique genetic context observed. The ZorA and ZorB sequences from the representative Zorya systems were then clustered using MMseqs2 with the options --min-seq-id 0.3 --coverage 0.95. The structures were predicted for each MMseqs2 cluster representative of each ZorA and ZorB family using ColabFold (v.1.5.2)⁷¹ with the options --num-recycle 3 --num-models 1 --model-type auto --amber --use-gpu-relax. Structure predictions were run as ZorA₅ZorB₂ multimers. The resulting structures were inspected manually (using PyMOL, v.2.5.4)⁵² to identify the start of the ZorA tail. The rest of the sequences in each cluster were aligned to the representative sequence using MUSCLE (v.5.1)³⁷ using the Parallel Perturbed ProbCons algorithm (default) or the Super5 algorithm if the cluster contained more than 100 sequences. The start of the tail for the representative was used to infer the start of the tail for each other protein in the respective alignment.

Figure preparation

Structural figures were prepared using ChimeraX v.1.8⁴⁸, PyMOL v.2.5.4 and v.3.0.2⁵², GraphPad Prism 9 or GraphPad Prism 10 (one-way ANOVA followed by Dunnett’s multiple-comparisons test)³³ and Adobe Illustrator⁷². The ion-permeation pathway shown in ZorAB was analysed using MOLEonline⁷³. The hydrophobicity and polarity of the ZorAB tail was calculated using MOLEonline⁷³. The electrostatic potential maps were calculated using the APBS⁷⁴ electrostatic Plugin integrated inside PyMOL. Full gel/blot images for all relevant figures are provided in Supplementary Fig. 1.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.