Introduction

Human enteric pathogenic bacteria, including Salmonella species, pose a significant global health burden, contributing to substantial morbidity and mortality despite the availability of antibiotics1. Salmonella is a Gram-negative bacterial pathogen responsible for a wide range of diseases in diverse animal hosts2 and is one of the four leading causes of diarrheal diseases globally3. Salmonella enterica serovar Typhi (S. typhi) infects only humans, while nontyphoidal Salmonella (NTS) are widely found in animal hosts4. Annually, S. typhi is estimated to cause 16–33 million infections annually, resulting in 500,000–600,000 deaths, while NTS infections account for approximately 90 million cases and 155,000 deaths globally4. Notably, the actual incidence of salmonellosis is likely significantly underestimated, with studies suggesting that for every reported case, there are approximately seven unreported cases in the community5.

The extensive use of antibiotics in healthcare and agriculture has accelerated the emergence of multidrug-resistant (MDR) bacterial pathogens6, which are predicted to cause 169 million deaths globally from 2025 to 20507. The 2016 European Union summary report on antimicrobial resistance highlighted very high to extremely high MDR levels in certain Salmonella serovars, including S. infantis, S. enteritidis, S. kentucky, and S. derby8. Throughout the farm-to-fork pathway, opportunities for Salmonella contamination persist9. Upon ingestion, Salmonella survives gastric acidity and overcomes host intestinal defenses, gaining access to the epithelium2. Once inside host cells, the bacteria replicate within the cytoplasm and within specialized compartments known as Salmonella-containing vacuoles10. This intracellular localization protects the bacteria from antibiotics and host immune responses11, complicating treatment. Additionally, antibiotics disrupt normal gut microflora and can lead to antibiotic-associated diarrhea, further challenging conventional therapeutic approaches12.

Phage therapy has gained attention as a promising alternative therapeutic strategy again for its ability to target drug-resistant bacteria, disrupt biofilms, and access intracellular pathogens13. Phages, the most diversified and dominant members of the gut virome, play a crucial role in shaping gut microbial communities and influencing host health14. Engineering mutations in phage host-range-determining regions to counteract bacterial resistance mutations has been shown to effectively prevent MDR Escherichia coli infection15. Furthermore, certain phages can restore bacterial antibiotic susceptibility and reduce virulence by inducing detrimental surface mutations in the host bacteria, thereby enhancing therapeutic outcomes16. For example, phage JNwz02 has been shown to restore antibiotic sensitivity in MDR S. anatum 2089b under selective pressure17.

Tailed phages use specialized tails for host recognition and genome delivery18, whereas tailless microviruses, small icosahedral phages with T = 1 symmetry, employ distinct, less understood mechanisms. Microviruses, which are classified as Microviridae family, are the most abundant single-stranded DNA (ssDNA) phages on Earth and an important component of the human gut virome19. The Microviridae family is divided into subfamilies Bullavirinae and Gokushovirinae. Atomic structures of Bullavirinae members, including φX17420, G421, and α322, have been determined through X-ray crystallography, revealing capsid protein F, spike protein G, and DNA binding protein J. These phages are hypothesized to employ spike protein G to interact with lipopolysaccharides (LPS) on host membranes, subsequently triggering the ejection of their small genome through a proposed coiled-coil tube formed by pilot protein H (hereafter “H tube”)23,24. However, the events occurring within the host cell during this process remain unexplored.

Unlike Bullavirinae, which require two scaffolding proteins to assemble, Gokushovirinae members are characterized by a single scaffolding protein system and the absence of major spike proteins25. Recent cryogenic electron microscopy (cryo-EM) studies on phages such as φEC609826, SpV427, and Ebor28 have revealed their unique features. Compared to Gokushovirinae, which typically have a longer lifecycle29, Bullavirinae demonstrate rapid progeny production, with infectious particles detectable as early as 5 min after host penetration by φX174-like viruses30. Minor variations in Bullavirinae structural proteins can significantly influence viral assembly, fitness, and genome evolution, driven by their small genome size and rapid lytic lifestyle25,31,32,33. These characteristics make the Bullavirinae subfamily an excellent model for studying virus evolution and infection mechanisms, offering valuable insights for future advancements in virus engineering and therapeutic strategies.

This study focuses on two Bullavirinae phages infecting S. enterica, named PJNS001 and PJNS002, which have not been previously characterized to the best of our knowledge. We determined the high-resolution cryo-EM structures and identified their LPS recognition patterns. The structural information suggested the key residues potentially involved in viral assembly and stability. Additionally, preliminary cryogenic electron tomography (cryo-ET) observations provided contextual views of phage-infected cells, offering a complementary perspective on phage-host interactions.

Results

Phage origin and receptor specificity

LPS cover the outer membrane of Gram-negative bacteria and play a central role in interactions with the environment. LPS are tripartite molecules composed of lipid A, a core oligosaccharide, and a variable O-antigen (Fig. 1a). Lipid A anchors LPS to the outer membrane and is responsible for its endotoxic activity34. It is linked to an inner core of unusual sugars and an outer core of more common sugars. In most strains, an O-antigen polymer extends from the core and contributes to surface diversity. Genes involved in LPS biosynthesis are organized in the rfa locus. rfaL encodes the ligase that links the O-antigen to the outer core, while rfaG, rfaI, and rfaJ encode glucosyltransferases involved in constructing the outer core35. Mutations in this locus can lead to truncated LPS and altered phage susceptibility.

Fig. 1: Phage origin, receptor specificity, and phylogenetic tree of PJNS001 and PJNS002.
figure 1

a Schematic representation of LPS structural mutations recognized by PJNS001 and PJNS002. Sugar residues are labeled using standard abbreviations: Man, mannose; Rha, rhamnose; Gal, galactose; Glc, glucose; GlcNAc, N-acetylglucosamine; Hep, L-glycero-D-manno-heptose; Kdo, 2-keto-3-deoxyoctonate; EtNP, ethanolamine phosphate; P, phosphate. Dashed arrows indicate the specific sugar positions added by enzymes encoded by the rfaL, rfaJ, rfaI, and rfaG genes. Hollow arrows indicate the LPS truncation points associated with susceptibility to PJNS001 (ΔrfaL) and PJNS002 (ΔrfaJ), showing that each phage targets distinct regions of the LPS core. b Phage sensitivity assay for PJNS001 and PJNS002 with different mutations. Infection specificity of phages PJNS001 and PJNS002 toward distinct receptor-deficient strains was determined via spotting assays. c Phylogenetic tree of PJNS001 and PJNS002 (red star). Phylogenetic analysis was performed using VipTree software. A total of 155 phages representing distinct viral families were analyzed, including Inoviridae, Microviridae, Plectroviridae, Pleolipoviridae, and Paulinoviridae.

Full size image

Previous studies identified rfaL as the receptor for Salmonella phage vB_SalS_JNS02 (GenBank no. PP189839)36, and demonstrated that this phage is unable to infect host bacteria lacking rfaL. To investigate alternative phage-host interactions, we screened using an rfaL-deficient ATCC14028 host strain and isolated several candidates. Notably, phage PJNS001—failing to lyse the wild-type strain—lysed only the rfaL-deficient host and lost infectivity upon further deletion of rfaJ. In contrast, phage PJNS002, isolated using an rfaJ-deficient host, lysed both rfaL- and rfaJ-deficient hosts but not wild-type or rfaI-deficient strains (Fig. 1b).

Whole-genome sequencing revealed that PJNS001 (5,386 bp; 45% GC) and PJNS002 (5486 bp; 46.1% GC) are ssDNA phages belonging to the genus Sinsheimervirus in the family Microviridae (Fig. 1c). Their genomes have no virulence or antibiotic resistance genes, nor any genes associated with transduction of bacterial DNA or toxin production. Members of Sinsheimervirus are known to be lytic, further supporting their potential safety for therapeutic use37.

Overall structures of PJNS001 and PJNS002

Negative staining was employed to assess sample purity, concentration, and heterogeneity38. The target virus particles were identified as pentagonal with visible spikes (Supplementary Fig. 1a). These particles were classified into two groups: full and empty. The empty particles likely underwent an irreversible conformational change39, leading to the entry of uranyl acetate.

We performed single-particle analysis (SPA) to reconstruct the density maps of PJNS001 and PJNS002 at resolutions of 2.68 Å and 2.59 Å, respectively, applied with icosahedral symmetry (Fig. 2, Supplementary Fig. 1b-c, Supplementary Table 1). These high-resolution density maps enabled the construction of atomic models for both viruses. Each icosahedral asymmetric unit (ASU) contains nearly full-length capsid protein F (427 residues for PJNS001 and 429 residues for PJNS002), spike protein G (175 residues), and DNA binding protein J (38 residues for PJNS001 and 26 residues for PJNS002) (Fig. 2b, 2d). Additionally, nine nucleotides were identified in the PJNS002 ASU.

Fig. 2: Cryo-EM structures of PJNS001 and PJNS002.
figure 2

a Cryo-EM electron density map of PJNS001 at a contour level of 1.0. Pentagons, triangles, and diamonds mark fivefold, threefold, and twofold symmetry axes, respectively. Capsid protein F, spike protein G, and DNA-binding protein J are colored in cornflower blue, light sea green, and gold, respectively. Due to the internal location of protein J, its density is not visible in the overall map. To visualize interactions with one protein G copy, proteins surrounding the fivefold axis are differentially colored (details in Fig. 3a). Insets show representative interacting residues between capsid subunits near each axis. b Cartoon representation of the ASU within the dashed triangle indicated in panel a, colored as cornflower blue (F), light sea green (G), and gold (J) to indicate the generality. c Cryo-EM electron density map of PJNS002 at a contour level of 0.65. Protein F, G, and J are shown in deep sky blue, turquoise, and lime, respectively, with subunits near the fivefold axis colored as Fig. 3b. d Cartoon representation of the ASU in panel a.

Full size image

Structurally, the regions spanning residues 74 (His for PJNS001 while Tyr for PJNS002) to Gly89 and Gln214 to Arg234 contribute to the two-fold interface. In addition, the region from Pro173 to Met213 forms the 3-fold interface, while the region from Ser246 to Val265 forms the 5-fold interface. These regions provide essential flexibility and rigidity and are highly conserved across the Bullavirinae subfamily (Supplementary Fig. 2a). For these two viruses, the maximum distance between spike proteins along the five-fold axes is approximately 35 nm, while the corresponding distance for capsid proteins is ~26 nm. The capsid diameters measured ~29 nm along the three-fold axes and ~23 nm along the 2-fold axes.

Both PJNS001 and PJNS002 exhibit slightly larger dimensions than previously determined Bullavirinae structures20,21,22. Earlier structures reported a maximum diameter of ~34 nm, with capsid diameters of ~25 nm along the five-fold axes and ~28 nm along the three-fold axes. Superposition with φX174 shows that the spike proteins in PJNS001 and PJNS002 are positioned inward by ~1 Å (Supplementary Table 2), suggesting that the observed size differences primarily result from a subtle expansion of the capsid. Given that only eight amino acid differences exist between the PJNS001 capsid protein and that of φX174 (Supplementary Fig. 1d), the size difference is unlikely to arise from significant native conformation changes. Instead, differences in experimental conditions, such as the longer purification times and crystallographic methods used in previous studies, might account for the more compact structures observed earlier.

Capsid protein assembly

Both capsid proteins lack the N-terminal Met, a characteristic shared across all resolved Microviridae structures20,21,22,26,27,28. The structural constraint strongly precludes the inclusion of an additional Met at the trimeric interface (Supplementary Fig. 1e). Mass spectrometry analysis further reveals that the PJNS002 capsid protein undergoes N-terminal Met excision and acetylation (Supplementary Fig. 1f). These modifications are mediated sequentially by Met aminopeptidase (MetAP) and N-acetyltransferase (Nat)40. Given the conservation of MetAP41 and Nat42 across all domains of life, similar post-translational modifications are likely common within the Microviridae family. These modifications may ensure proper capsid assembly or enhance phage proliferation, as proposed in other phages, such as Salmonella phage SPN3US43 and Streptococcus thermophilus phage DT144.

While the flexible apically distal head domain of the three-fold protrusion in the Gokushovirinae subfamily remains unresolved, capsid proteins across the Microviridae family share a conserved jelly-roll motif composed of an eight-stranded antiparallel β-barrel (Supplementary Fig. 3a). The EF and HI insertions form most of the capsid protein’s outer surface, while N-terminal or C-terminal extensions in the Gokushovirinae subfamily may enhance stability or curvature26,28. Despite the highest amino acid sequence identity of structural proteins between PJNS002 and other Bullavirinae members being 76.83% (capsid protein) (Supplementary Table 3), the Cα-distance between PJNS001 and PJNS002 capsid proteins is only 0.487 Å, highlighting the conservation of capsid proteins across the Bullavirinae subfamily.

Larger Cα-distances are observed when comparing PJNS001 and PJNS002 to other Bullavirinae members. These differences are primarily attributed to a subtle, overall expansion that influence surface area and intermolecular interactions (Supplementary Table 4). Such expansions may reflect minor structural adaptations across the Bullavirinae subfamily while preserving the essential architectural features of the capsid.

Pentameric interfaces analysis

While the β-barrel structure of the spike proteins is conserved across the Bullavirinae subfamily, notable variability exists at the pentameric contact sites. For description, protein copies, numbered sequentially from G1 to G5, were defined in a clockwise direction in the reconstructed pentamer. In PJNS001, G1 forms nine hydrogen bonds with G2, one with G3, and five with F proteins (Fig. 3a). In PJNS002, G1 forms twelve hydrogen bonds with G2 and two with F proteins (Fig. 3b). Conserved hydrogen bonds, such as G1Met1-G2Asn10, G1Phe2-G2Phe5, G1Phe2-G2Ser7, G1Gln3-F4Gln255, G1Ser15-G2Val109 (PJNS001)/G2Asp109 (PJNS002), participate in assembly.

Fig. 3: Inter-subunit interactions surrounding the fivefold axis of PJNS001 and PJNS002.
figure 3

a Fivefold symmetric assembly of PJNS001, shown with 40% surface transparency. Subunits are colored consistently with Fig. 2a: protein G2, G3, F2, F3, F4, and F5 are shown in tomato, cyan, teal, olive, magenta, and pale green, respectively. Other subunits are shown in light sea green (G), cornflower blue (F), and gold (J). Dashed black lines indicate hydrogen bonds. Three regions are boxed: Inset 1 highlights the interaction interface between G1 and G2; Inset 2 shows the interactions surrounding the N-terminus of G1; Inset 3 displays a cross-sectional electrostatic surface of the G pentamer. b Fivefold symmetric assembly of PJNS002, with 40% transparency. Subunits are colored as in Fig. 2c: protein G2, F2, F4, and F5 are colored chocolate, dark khaki, hot pink, and aquamarine, respectively. Other subunits follow the scheme of turquoise (G), deep sky blue (F), and lime (J).

Full size image

In PJNS002, an additional hydrogen bond between G1Ser15 and G2Leu110 may compensate for the weaker interaction between G1Ser15 and G2Asp109. Similarly, the interaction between G1Thr171 and F2Ser399 in PJNS002 may be equivalent to that of G1Glu167-F2Glu367 and G1Lys175-F2Gln396 in PJNS001. G1Met1 interacts with G3Thr128 in PJNS001, but with G2Ile170 in PJNS002. These flexible interactions likely represent structural adaptations and serve to prevent mixed spike pentamer formation in co-infected cells25.

The center of the spike protein pentamer contains a hydrophilic channel formed by amino acids with varying charges. This channel entrance is characterized by twisting and diameter variations influenced by residues with differing hydrophilicity (Supplementary Fig. 4). Here, PJNS001 features negatively charged Asp117, whereas PJNS002 presents uncharged Ser118 and Asn119. Below this, the positively charged Arg115 in PJNS001 contrasts with the uncharged Thr116 in PJNS002. These charge differences result in variations in channel surface shape, which may influence pore selectivity or hinder hybrid pentamer formation. Further below, positively charged Arg and Lys residues form a spacious area. At the pentamer bottom, hydrophilic residues, including Thr4 in PJNS001 and negatively charged Glu4 in PJNS002, form a larger cavity (Supplementary Fig. 5). These charged polar residues may modulate pentamer functionality.

At the pentameric interface of Bullavirinae capsid proteins, the conserved DGTDQ motif (residues 251–255) forms a central pore, with T253 contributing to a diameter of ~4 Å (Supplementary Fig. 6a–c). In contrast, corresponding regions in the Gokushovirinae subfamily are more variable, with different motifs and pore sizes (Supplementary Fig. 6d–f). These structural differences may reflect functional adaptations across subfamilies, with the conserved architecture in Bullavirinae likely supporting specific roles in viral assembly and stability.

Distribution of DNA binding proteins

Inside the capsid, DNA binding proteins, characterized by their strong alkalinity, insert into the capsid groove, counterbalancing the slightly alkaline inner surface of the capsid proteins (Fig. 4a, b, Supplementary Table 5). In PJNS001, the DNA binding protein features an extended N-terminus with 12 additional residues that traverse the adjacent capsid protein’s β-barrel. Notably, the resolution of the DNA binding protein reveals significant differences between the Bullavirinae and Gokushovirinae subfamilies. In the Gokushovirinae subfamily, the DNA binding protein is often resolved with fewer residues or remains unresolved (Supplementary Fig. 3b). Taking φEC6098 as an example, the absence of N-terminal density of DNA binding protein is attributed to the high N-terminal flexibility caused by its small genome size (~4.5 kb)26. However, SPV4, which has a similarly small genome (~4.4 kb), resolves part of the N-terminal structure27, while Ebor, with a larger genome (~6.6 kb), fails to resolve the DNA binding protein at all28. These findings suggest that genome size alone does not determine DNA binding protein resolution. Instead, particle integrity appears to be a critical factor, as negative staining revealed increased numbers of empty particles in our samples, consistent with a large number of empty particles observed in SPV4 and Ebor samples.

Fig. 4: Electrostatic potential and interaction details of protein J in PJNS001 and PJNS002.
figure 4

a Electrostatic potential surface of the PJNS001 fivefold assembly, viewed from the interior. The close-up is colored according to the scheme in Fig. 3a. b Electrostatic potential surface of PJNS002 fivefold assembly, shown from the interior. Single-stranded DNA (ssDNA) is colored as crimson in the close-up. c Interaction analysis between protein J and F in PJNS001. The density of protein J is displayed at 0.8 threshold with 40% transparency. Three boxed regions are magnified to show representative interacting residues, with hydrogen bonds indicated by dashed lines. d Interaction analysis between protein J and both protein F and ssDNA in PJNS002.

Full size image

In PJNS001, J1 forms six hydrogen bonds with F5 and eight with F1; in PJNS002, it forms seven hydrogen bonds with F5 and three with F1 (Fig. 4c, d). The residues Asp14, Asp62, and Arg215 on the inner surface of the capsid protein are positioned at the β-barrel, HI insertion, and EF insertion, respectively. Conserved hydrogen bonds between these regions and DNA binding protein likely contribute to capsid stability, although the specific interacting residues differ between the two viruses. Additionally, nucleotides in PJNS002 are stabilized through interactions with long side-chain residues of DNA binding protein, including Lys3, Arg7, Arg17, and Tyr20. These interactions suggest a crucial role for DNA binding protein in maintaining genome stability.

Thermodynamic analysis

To better understand the molecular basis of phage assembly, we performed a thermodynamic analysis of macromolecular interfaces using PDBePISA, an online tool that evaluates the structural and chemical properties of macromolecular surfaces and interfaces and predicts the biological relevance of observed contacts45. The analysis showed that the assembly of PJNS001 and PJNS002 is thermodynamically characterized by facilitated assembly but reduced stability. This is reflected by greater free energy release during assembly and less energy required for dissociation (Supplementary Table 6).

In the Bullavirinae subfamily, capsid assembly is primarily stabilized by the pentameric interface, followed by dimeric and trimeric interfaces (Supplementary Table 7). In contrast, capsid assembly mechanisms within the Gokushovirinae subfamily are more diverse. For instance, in φEC6098, the N-terminal extension contributes to the stability of the F protein trimer’s bottom region, whereas in Ebor, the C-terminal extension stabilizes the F protein dimer (Supplementary Fig. 3a, Supplementary Table 8). These differences highlight the evolutionary flexibility of assembly mechanisms across Microviridae, potentially reflecting structural adaptations to distinct functional or environmental constraints.

Interactions between spike and capsid proteins show minor thermal changes and large P values, indicating weak intrinsic affinity and the need for external scaffolding proteins during assembly. Additionally, although the DNA binding proteins in Bullavirinae vary naturally in length (Supplementary Fig. 2b), their conserved C-terminal regions form similarly sized interface areas with the overlying capsid proteins (Fig. 4a, b, Supplementary Table 7), underscoring the structural importance of the C-terminus in maintaining interface stability.

LPS recognition of PJNS002

To investigate the substrate recognition mechanism, we expressed and purified the spike protein G of PJNS002 (Supplementary Fig. 7a). Size-exclusion chromatography showed that protein G predominantly formed pentamers with high purity, and SDS-PAGE indicated that these pentamers dissociated into monomer-sized units after denaturation. Due to the limited solubility of wild-type LPS (up to ~1 mM), 0.025% DDM was added to enhance solubility and ensure consistent LPS concentrations during binding assays.

Biolayer interferometry (BLI) exhibited a biphasic curve, characterized by rapid association and dissociation phases, followed by signal stabilization above baseline (Supplementary Fig. 7b). This suggests distinct binding events rather than a simple 1:1 interaction. Kinetic modeling further supported a 2:1 heterogeneous binding model, which fit the data markedly better. Given this and the structural heterogeneity of LPS—with various truncated forms differentially exposing or obscuring potential binding epitopes—we adopted the 2:1 model to interpret the data. The derived affinity constants, KD1 = 19.4 ± 31.1 mM and KD2 = 35.3 ± 2.13 μM, indicate low affinity, consistent with the fact that PJNS002 is more inclined to bind LPS from host strains with rfaJ mutations. Such weak, transient interactions may reflect the dynamic nature of phage adsorption to structurally diverse LPS landscapes on host cells46.

To address the orientation preference in the sample of protein G-LPS complex, we implemented stage-tilt data collection (Supplementary Fig. 7c)47 and 2D classification rebalancing, but still yielded a 3.95-Å map with overfitting artifacts (Supplementary Fig. 7d, EMD-62020). Although an extra density adhering the protein G may represent LPS, the precise structural details could not be well resolved.

Discussion

Phages and bacteria have co-evolved for billions of years, engaging in a constant evolutionary arms race that drives their genetic and structural diversification48. The rising prevalence of MDR Salmonella has intensified interest in phages as potential antibacterial agents. In this study, we employed structural and biochemical analyses to characterize Bullavirinae phages PJNS001 and PJNS002, providing important insights into receptor recognition and assembly mechanisms within the Microviridae family.

We encountered significant challenges in resolving the structures of the G protein complexed with LPS. To gain further insights into the infection process, we utilized Cryo-ET, an advanced technique in Cryo-EM field capable of visualizing heterogeneous biological structures in their near-native state by acquiring tilted series for tomogram reconstruction, which is well-suited to studying dynamic interactions between phages and host bacteria in situ.

The invasion process of φX174 involves three major stages: attachment to the host surface, eclipse (losing part of the genome), and DNA penetration into the cytoplasm39. To visualize the eclipse stage, during which the H tube is expected to form, we incubated phages with host bacteria at 36°C in PBS for 10 min before vitrification (Supplementary Fig. 8a). Tomographic reconstructions revealed host cells’ inner and outer membranes, with phage particles randomly attached to the outer membrane (Supplementary Fig. 8b). Notably, some bacterial cells showed signs of early lysis (Supplementary Fig. 8c), indicative of phage activity.

Sub-tomogram averaging (STA) was then performed on manually picked viral particles, but reconstructions of PJNS001 (EMD-62021) and PJNS002 (EMD-62022) with C5 symmetry did not reveal the presence of the H tube. Low-resolution maps showed thinning of capsid near the five-fold axis on the attachment-facing side (Supplementary Fig. 8d), suggesting structural remodeling at the portal region.

We measured distances between the host’s inner and outer membranes at all viral attachment sites (Supplementary Fig. 8e), calculating the mean distances of 442.1 Å for PJNS001 and 425.2 Å for PJNS002, with statistically significant differences in median distances (Supplementary Table 9). Further experiments under varying incubation durations (5, 10, and 20 min) with PJNS002 yielded corresponding median intermembrane distances of 285.8 Å, 348.7 Å, and 276.3 Å, respectively (Supplementary Fig. 8f), implying that this distance may fluctuate with infection progression.

The absence of visible H tubes may reflect their transient existence or limitations in current cryo-ET sample preparation and resolution. SPA of phage T7 has revealed tunnel-like periplasmic structures during DNA ejection49, but in situ reconstructions remain limited to ~4 nm resolution50, underscoring the challenges of visualizing such ephemeral events. Similar challenges were reported for φX174-like phage ST-1, in which periplasm-penetrating tubes were only occasionally observed23. Additionally, the rapid ejection of highly negatively charged viral DNA would exacerbate electron scattering, thereby obscuring the H tube density in reconstructions23,24.

Future technical advances, such as time-resolved cryo-EM, fluorescent labeling, and liquid-phase electron microscopy, hold promise for resolving virus-host interactions in real time51. Such interdisciplinary approaches may yield deeper insights into Microviridae infection mechanisms and support the rational design of improved phage therapeutics.

The evolutionary dynamics and ecological adaptations of Microviridae phages remain poorly understood. Phylogenetic analysis of their capsid proteins suggests PJNS001 may represent a more ancestral position relative to φX17452. Among eight amino acid substitutions in PJNS001 capsid proteins (Supplementary Fig. 1d), six match putative ancestral residues, while the other two (T93A, V319R) are positionally conserved but differ in substitution type.

Despite extensive nucleotide changes between φX174 and PJNS001, only 30 amino acid substitutions occurred, alongside 189 synonymous mutations53. A notable substitution (L243F in the capsid protein) preserves overall charge and hydrophobicity while introducing significant steric and electrostatic alterations52, which could contribute to thermal adaptation54,55. Synonymous mutations could influence viral assembly by modifying internal capsid pressure56 or codon usage bias57.

The high mutation rates in the spike proteins of Bullavirinae phages may reflect adaptation to diverse Salmonella serotypes (Supplementary Fig. 2c), enhancing their infectivity spectrum. However, in the ongoing co-evolutionary arms race, bacteria often retain a selective advantage. This is because modifying surface receptors is typically more accessible for bacteria than for phages to evolve new receptor-binding capabilities58. For example, Salmonella can utilize bistable epigenetic switches to diversify surface receptor expression, increasing the probability of survival under phage predation59. Thus, while Bullavirinae phages show potent lytic activity under laboratory conditions, their practical deployment is limited by host diversity and environmental constraints such as gastrointestinal acidity.

To overcome these limitations, constructing diverse phage libraries targeting multiple LPS variants may enhance host coverage and clinical applicability. While no specific residues of spike proteins were conclusively validated as engineering targets in this work, the putative LPS densities offer initial insights into the interaction interface and may inform future rational design strategies to expand host range. Such efforts could pave the way for the development of next-generation phage therapeutics tailored to combat MDR Salmonella infections.

Methods

Isolation, amplification, and purification

Phages were isolated from environmental samples using a direct isolation method, followed by plaque purification via the conventional double-agar overlay technique60. Briefly, 50 mL wastewater was collected from Jinan, China, and then filtered through a 0.22 μm membrane to remove bacteria and debris. Subsequently, 200 μL of filtered wastewater was mixed with 200 μL of the indicator strain Salmonella ATCC 14028-ΔrfaL (for PJNS001) or ATCC 14028-ΔrfaJ (for PJNS002). Both strains harbor the Kan resistance gene. The mixtures were added to 4 mL soft LB agar (0.75%) and overlaid onto prepared LB agar plates (1.5%) with a final concentration of 50 mg/L Kanamycin. Each phage was cultured on 10 individual plates. All plaques were harvested and pooled into 50 mL SM buffer (comprising 5.8 g NaCl, 2.0 g MgSO4·7H2O, 50 mL 1 M Tris-HCl [pH 7.4], and 50 mL 2% [w/v] gelatin per liter), then incubated overnight at 4 °C with gentle shaking (180 rpm) to elute the phage particles.

Cell debris was removed by centrifugation at 2000 × g for 10 min. The filtered supernatant through a 0.22 μm filter (Sangon) was further purified by centrifugation at 222,200 × g for 2 h. The resulting pellet was resuspended in 15 mL PBS buffer (Sangon) for 2 h and concentrated to 400 μL using a 100 kDa Ultra Centrifugal Filter (Merck). A 15–45% [w/v] sucrose gradient in PBS was prepared using Gradient Master 108 (Biocomp), at an angle of 81.5°, speed 17, for 1 min 53 s in 12.5 mL open-top ultracentrifuge tubes. The concentrated phage solution was gently layered atop the gradient, which was then centrifuged in a SW41 rotor (Beckman) at 154,300 × g for 2 h. The gradient was fractionated into 25 sequential 500 μL fractions from top to bottom. Virion bands were consistently found in fractions 15~20, as confirmed by negative staining electron microscopy.

Spotting assay

To determine the infection specificity of phages PJNS001 and PJNS002 toward different receptor-deficient strains, spotting assays were performed61. Briefly, 200 μL of mid-exponential phase bacterial cultures were inoculated into 4 mL soft LB agar (0.75%) and overlaid onto solidified LB agar plates (1.5%). Subsequently, Aliquots (2 μL) of phage PJNS001 or PJNS002 lysates at varying titers were individually spotted onto designated areas of the agar plates. The plates were incubated at 37 °C for 12 h, and bacterial lysis was assessed through plaque formation.

Genome sequencing and analysis

The genomic DNA of phages PJNS001 and PJNS002 was extracted using the phenol-chloroform method62. Sequencing libraries were prepared in accordance with the manufacturer’s instructions utilizing the NEBNext Ultra II FS DNA Library Prep Kit (New England Biolabs, NEB)63. Sequencing was conducted on an Illumina NovaSeq 6000 platform, employing paired-end reads of 2 × 150 bp. The quality of the sequencing data was assessed using FastQC v0.11.564. The paired-end reads were subsequently filtered and assembled using SPAdes v3.13.0 with default parameters65. The complete genomes of phages PJNS001 and PJNS002 were annotated using the online server RAST66. Protein functions were validated through the local alignment search tool BLASTp server (https://blast.ncbi.nlm.nih.gov/). Additionally, tRNAscan-SE v2.0 (http://lowelab.ucsc.edu/tRNAscan-SE2/)67, ResFinder (https://cge.cbs.dtu.dk/services/ResFinder/)68, and VirulenceFinder (https://cge.cbs.dtu.dk/services/VirulenceFinder/)68 were utilized to identify the presence of transfer RNA (tRNA), resistance genes, and virulence factors, respectively. Viptree software (https://www.genome.jp/viptree/) was employed for phage PJNS001 and PJNS002 phylogenetic analysis.

Expression and purification of PJNS002 protein G

The gene encoding PJNS002 protein G was codon-optimized for E. coli, synthesized and inserted into the pET28a(+) vector (Genewiz). The E. coli BL21 (DE3) clone containing the plasmid was selected on an LB agar plate with kanamycin and subsequently amplified. Following cultivation in LB medium with kanamycin at 37°C and shaking at 220 rpm for 5 h, Isopropyl β-D-1-thiogalactopyranoside (IPTG) was added to the culture to a final concentration of 0.2 mM. The culture was then incubated at 16°C overnight to express the protein. Bacterial cells were harvested by centrifugation at 4000 × g for 20 min. The resulting pellet was resuspended in lysis buffer (20 mM Tris pH 8.0, 10% [w/v] glycerol, 300 mM NaCl). The bacteria were lysed using a Microfluidic Ultra-High-Pressure Homogenizer (Litu) and clarified by centrifugation at 40,000 × g for 45 min. The supernatant was loaded onto a Ni-NTA affinity column at 4°C for 2 h. After being washed with lysis buffer containing 100 mM imidazole, the proteins were eluted using lysis buffer containing 500 mM imidazole. The eluted proteins were concentrated using a 30 kDa Ultra Centrifugal Filter (Merck) and further purified by Size Exclusion Chromatography using a Superdex 200 Increase 10/300 column (Cytiva) with buffer (20 mM Tris pH 8.0, 150 mM NaCl).

Negative staining electron microscopy

For phage samples with sucrose, 3 μL of the sample was loaded on a freshly glow-discharged (Ted Pella) carbon-coated copper grid (EMCN) and manually blotted with filter paper. 3 μL PBS was applied to the grid and blotted away three times to reduce the extra sucrose. Subsequently, 3 μL of 1% uranyl acetate solution was applied and blotted away three times. The grid was left to dry and imaged using a Talos L120C electron microscope (Thermo Fisher Scientific) operated at 120 kV, typically at 73,000× magnification for image acquisition. For protein samples, the PBS rinse steps were skipped.

Cryo sample preparation and data acquisition

For samples of PJNS001 and PJNS002, virus-containing fractions identified by negative staining EM were pooled and concentrated from ~3 mL to ~500 μL using a 15 mL 100 kDa centrifugal filter (Merck). To remove sucrose, PBS without sucrose was added to a final volume of 5 mL, gently mixed for 10 min, and reconcentrated to 500 μL. This dilution-concentration cycle was repeated three times in total, followed by final concentration to ~20 μL using a 1 mL 100 kDa centrifugal filter.

3 μL of the sample was applied on a freshly H2/O2 glow-discharged (Gatan) R1.2/1.3 holey copper grid (Quantifoil). The grid was blotted for 3 s with a blot force of -1 at 4°C and 100% humidity, then plunge-frozen in liquid ethane using a Vitrobot IV (Thermo Fisher Scientific). Forty consecutive frames were recorded using a Titan Krios G3 TEM (Thermo Fisher Scientific) equipped with a K3 detector (Gatan) operated at 300 kV. All frames were automatically collected in counting mode using SerialEM 4.1.969. The data acquisition parameters included a defocus range of 0.8 to 2.0 μm, a pixel size of 0.53 Å in super-resolution mode and a total electron dose of 50 e-/ Å2 for each frame stack. In total, 4808 frame stacks were collected for PJNS001, and 3827 frame stacks for PJNS002.

For the sample of spike protein G interacting with LPS, 5 μL of 0.25 mg/mL PJNS002 protein G was mixed with 1 μL of 5 mg/mL S. enterica LPS (Merck) in the SEC buffer. Then, 3 μL of the mixture was applied to an R1.2/1.3 holey gold grid and processed following the plunge-freeze steps described above. However, frame stacks were collected with different parameters and at varying stage tilt angles. For each frame stack, the pixel size is 0.41 Å in super-resolution mode and the total electron dose is 60 e-/ Å2. A total of 712, 1,940 and 863 frame stacks were collected at tilt angles of 0°, 15° and 30° with defocus ranges of 0.8~2.0 μm, respectively.

Image processing and model building

For datasets of PJNS001 and PJNS002, frame stacks were motion-corrected and dose-weighted with 2× binning using MotionCor270 in Relion 3.1.371. Subsequent processing steps were performed in CryoSPARC 4.0.172. After patch CTF estimation, particles were picked using the blob picker, extracted with a box size of 400 pixels and Fourier-cropped to a box size of 200 pixels. Following the initial round of 2D classification, classes with clear features were selected and re-extracted with 400 pixels. The accepted particles were then subjected to the second round of 2D classification. Ab initio reconstruction and non-uniform refinement were then performed using icosahedral symmetry, resulting in final maps with resolutions of 2.68 Å for PJNS001 (16,905 particles) and 2.59 Å for PJNS002 (86,486 particles).

For the dataset of protein G, patch motion correction with 1/2 F-Crop factor, patch CTF estimation and following steps were performed in CryoSPARC 4.5.1. For tilt angles of 0°, 15°, and 30°, 689, 1566, and 733 micrographs were retained, respectively. The protein G pentamer model was selected from the complete PJNS002 structure and converted to an electron density map using e2pdb2mrc.py in EMAN 2.073. This pentamer map was imported to create templates for the template picker. Multiple rounds of particle extraction and 2D classification were performed. The selected classes and accepted particles with a box size of 160 pixels were finally rebalanced into three superclasses. Ab initio reconstruction and non-uniform refinement were subsequently performed, resulting in a final map with a resolution of 3.95 Å (238,773 particles).

The structures of proteins F and G were predicted using AlphaFold274, then trimmed and docked into phage maps using Phenix 1.20.175. Gaps in the structures and protein J were manually built using Coot 0.9.876. Multiple rounds of real-space refinements and manual corrections were performed. The complete models were generated by docking 60 copies of the F-G-J monomer into the maps.

Cryo-ET sample preparation and data acquisition

For plunge-freezing of host bacteria invaded by phages, host bacteria were cultured in 5 mL of LB medium to an OD600 of 0.5~0.8. After centrifugation at 2000 ×g for 10 min, the pellet was resuspended in 50 μL of PBS. A mixture consisting of 2 μL of the bacteria sample, 2 μL of phage samples and 1 μL of 10 nm gold beads was incubated at 36°C for 10 min. Then the mixture was plunge-frozen using an R2/2 H2 gold grid following the previous parameters and procedures.

Tilt series were collected on the same Titan Krios G3 TEM equipped with a K3 detector, using the PACEtomo77 script in SerialEM. The acquisition parameters included a dose-symmetric scheme, a pixel size of 0.84 Å in super-resolution mode, a defocus of 4 μm, a tilt increment of 3°, a tilt range from -60° to 60° starting from 0°, 10 frames per tilt and a total dose of ~100 e-/Å2. A total of 18 and 32 tilt series were collected for bacteria invaded by PJNS001 and PJNS002, respectively.

To further investigate the invasion process, host bacteria were invaded by PJNS002 with a time gradient of 5, 10 and 20 min. The mixed solutions were incubated at 36°C for the specified time before plunge-freezing. Tilt series were collected with the same parameters as above. A total of 75, 157 and 120 tilt series were collected for bacteria invaded by PJNS002 for 5, 10 and 20 min, respectively.

Tomogram reconstruction and sub-tomogram averaging

Per-tilt averaged images were imported into IMOD 4.11.2478 after motion correction of tilt series with 2× binning using MotionCor2 in Relion 4.0.0. Most coarse alignments were done manually using Midas due to imprecise tracking during acquisition. Final alignments were achieved using linear interpolation and a 4× binning to enhance the signal. CTF correction utilized parameters estimated by Ctffind 4.1.1479 and dose weighting utilized statistics calculated from the mdoc files. Tomograms were then reconstructed by weighted back projection followed by a rotation around X axis. Denoised tomograms were generated using Isonet 0.280, after which particles were manually picked. The particle coordinates, converted to a 1× binning, were imported into Relion for subtomogram averaging. Initial models were generated from the particles. 3D classification and 3D auto-refinement were subsequently performed to generate the final maps. For the PJNS001 and PJNS002 samples incubated at 36 °C for 10 min, C5 symmetry was applied during data processing, resulting in final resolutions of 31 Å and 30 Å, respectively. For the PJNS002 samples incubated at 36 °C for 5, 10 and 20 min, C1 symmetry was applied, yielding a final resolution of 38.4 Å for all samples.

Biolayer interferometry binding assays

To remove Tris from the buffer, protein G was first desalted using a Zeba Spin Desalting Column (Thermo Fisher Scientific). Subsequently, biotinylation was carried out using the EZ-Link NHS-PEG4-Biotin reagent (Thermo Fisher Scientific), which attaches biotin to accessible primary amine groups (-NH₂). Biolayer interferometry experiments were conducted at 30°C with shaking (1000 rpm) using an Octet RED96 (ForteBio). Biotinylated protein G in kinetic buffer (PBS containing 0.025% w/v DDM and 1 mg/mL BSA) was immobilized onto streptavidin (SA) sensors (Sartorius) to a binding response of 4 nm, as measured by the shift in optical thickness. Following a 300-second baseline measurement in kinetic buffer, the biosensors were incubated with various concentrations of S. enterica LPS (50, 100, 200, and 400 nM) for 600 s and then dissociated in kinetic buffer for 300 s. The final binding curves were fitted using a 2:1 binding model, yielding kinetic values with χ² = 1.174 and R² = 0.9981.

Data analysis and visualization

Sequence alignments were performed using SnapGene 6.0.2. Most of the structural analyses and visualizations were conducted using ChimeraX 1.881. The segmented volume and manually docked phage volumes were rendered for visualization in Amira after segmentation operations such as membrane enhancement filter, interactive thresholding and removing small spots.

Statistics and reproducibility

Macromolecular interface exploration was carried out with PDBePISA45. Membrane distances were manually picked from invasion sites in tomograms. Original data for SEC and BLI were redrawn using GraphPad Prism 10.1.2. Statistical analysis, as indicated in each figure legend, was performed using ANOVA: Kruskal-Wallis test. Dunn’s multiple comparisons test was applied with significance levels indicated as * P < 0.05, **** P < 0.0001.

Reporting summary

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