The Y. pestis strains used in this study are described in Table 1 . Other bacterial strains and plasmids are described in Supplementary Table S1 , and the primers used in the study are described in Supplementary Table S2 . The phage Yep-phi used in this study can effectively lyse all tested Chinese isolates of Y. pestis but is unable to lyse other Yersinia species (Zhao et al., 2011). Y. pestis and Yep-phi cultures were incubated at 26°C and E. coli at 37 °C. Luria–Bertani (LB) media were used for bacterial liquid cultures. A soft agar medium was prepared by adding 0.4% (wt/vol) agar to liquid media. Plates were supplemented with ampicillin (Amp, 100 μg/ml), kanamycin (Kan, 100 μg/ml), spectinomycin (Spe, 100 μg/ml), or chloramphenicol (Cm, 34 μg/ml) when required.

Different titers (from 1.5 × 10⁸ plaque-forming unit [PFU] to 1.5 × 10⁴ PFU) of 0.5 ml strain 614F (10⁹ CFU/ml) and 5 ml Yep-phi were mixed and incubated at 26°C for 24h. The serial mixtures were separately plated on LB agar and incubated at 26°C for 24h. Colonies on plates were resistant to the corresponding titer of phage and were selected for passages. Bacterial stocks of every passage were stored for future analysis. Coculture passages were performed to obtain a fully resistant Yep-phi derivate of 614F (phage titer 1.5 × 10⁸ PFU). The genomes of resistant and wild-type 614F strains were sequenced and compared. The identified mutations were screened with mismatch–polymerase chain reaction (PCR) in all passage stocks to determine when these mutations occurred during the entire passage process ( Supplementary Table S2 ) (Wangkumhang et al., 2007).

Scarless genome editing was used to generate gene knockouts and point mutants in Y. pestis (Kim et al., 2014). In brief, selection cassettes containing Spe-resistant genes, flanked by FRT, I-SceI sites, and 120-bp homologous arms, were electroporated into 201/pREDTKI, which encodes meganuclease I-SceI under the control of the an hydrotetracycline-induced tetA promoter and λ-Red recombinase genes under the control of the arabinose-inducible araB promoter as well as a temperature-sensitive replication origin. The knockout strains were confirmed via PCR and sequencing. Using the primers listed in Supplementary Table S2 , the PCR fragment containing the SacI/HindIII site of mutant gene S56 was amplified and cloned into the donor plasmid pKSI-1, which with I-SceI recognition sites was digested with SacI and HindIII to produce pKSI-1_waaA*, pKSI-1_cmk*, and pKSI-1_{_} ail*. The donor plasmid was electroporated into knockout strains; transformants were restruck onto LB plates containing Kan, Spe, and Amp and grown overnight at 26°C. Restruck colonies were suspended in LB containing Kan, Spe, and Amp and grown overnight at 26°C. The colonies were repatched twice onto LB supplemented with 10 mmol L-arabinose for inducing lambda red recombinase and 20 mmol IPTG for inducing I-SceI expression to loop out the selection cassette and donor plasmid at 26°C. To verify the loss of the selection cassette, isolates were tested for growth on LB plates with or without Spe and Amp. The sensitive isolates were screened via PCR and sequencing to confirm mutagenesis. Successive passage was performed at 37°C to remove pREDTKI from the strain. The waaA*/cmk*/ail* mutant was constructed via suicide plasmid-mediated genome editing (Philippe et al., 2004). Homologous upstream and downstream fragments flanking the waaA* mutation were amplified from S56 using PCR. The fragments were cloned into the suicide plasmid pDS132, which was digested with SacI and SalI, and replicated in S17-1 λpir. The recombinant plasmids were purified and introduced in cmk*/ail* via conjugation. Plates containing Cm were screened after incubation at 37°C overnight. The clone was selected on a plate containing 7% sucrose following growth at 26°C for 4 days. We selected transformants that grew on plates but not on Cm plates. The strain containing waaA*/cmk*/ail* was detected via PCR and confirmed via sequencing.

The overnight grown strains were cultured to OD₆₂₀ of approximately 1.0, diluted 1:100 into 20 ml of LB, and cultured at 26°C and 200 rpm. OD₆₂₀ was determined every 2h using 300 μl of bacterial culture. The OD₆₂₀ values were used to draw growth curves. Data were obtained in triplicate, and experiments were repeated twice. Area under the curve analysis was used to quantify the differences in growth (Vornhagen et al., 2019).

The lysis activity of each phage strain was examined using a spot test on Y. pestis (Paul et al., 2011; Zhao et al., 2013). In brief, 100 ml of bacteria cultured to exponential phase (OD₆₂₀~1.0) were concentrated 10-fold via centrifugation and mixed with 3 ml of liquefied soft agar medium. The mixture was plated on LB plates to create double-layer agar plates. When the medium solidified, 3 μl of serial dilutions of Yep-phi were spotted on the plates. Plates were incubated at room temperature for 10 h and then examined and photographed.

For phage adsorption, approximately 8 × 10⁵ PFU of Yep-phi in 100 μl was mixed with 500 μl bacteria (OD₆₂₀~1.0). LB was used as a nonadsorbing control in each assay, and the phage titer in the control supernatant was set to 100%. The mixture was incubated at room temperature for 5 min and centrifuged at 16,000 × g for 3 min. Residual PFU percentage was calculated as described in a previous study (Kiljunen et al., 2011).

Statistical differences were determined using one-way analysis of variance (ANOVA) with three independent data sets. To reveal the temporal phage adsorption kinetics of all strains, residual PFU percentage was tested at different time points (2, 5, 10, and 15 min). Each assay was performed in triplicate and repeated twice. Statistical differences were determined using two-way RM ANOVA with Dunnett’s multiple comparisons to wild-type groups.

To determine the efficiency of plating (EOP) of Yep-phi on different strains, 300 μl of wild-type and mutant Y. pestis–cultured bacteria (OD₆₂₀~1.0) were mixed with 100 μl of Yep-phi in 3 ml of 0.4% soft agar and poured onto LB plates. The number of PFUs was counted after 24h–48h. Each strain was verified in triplicate. EOP was calculated as described (Hyman and Abedon, 2010).

LPS isolation was performed using the phenol–water extraction method (Zhang and Skurnik, 1994). In brief, 9 ml overnight bacterial cultures were collected via centrifugation and resuspended in 1.5 ml distilled water. The bacterial suspensions were incubated at 70°C for 1h and then mixed with water-saturated phenol (pH 4.0) reheated to 70°C for 10 min and then transferred to ice for cooling (< 10°C). Then, the cells were centrifuged at 2,000 × g for 20 min. The uppermost aqueous layer obtained was transferred to a new Eppendorf tube and 2 volumes of acetone were added to precipitate LPS. The LPS pellet was dissolved in 150 μl of water, of which 5 μl of LPS was analyzed via 15% sodium dodecyl sulfate-polyacrylamide gel electrophoresis and silver staining (Zhang and Skurnik, 1994).

The PCR-generated DNA fragment comprising waaA and waaA* was cloned into the BamHI/HindIII site of the pET32a vector. The plasmid was introduced into DH5α cells, and its DNA sequence was confirmed. The plasmid was extracted and introduced into E. coli BL21(DE3) cells. To induce the expression of WaaA and WaaA* in E. coli BL21(DE3) cells, 1 mM IPTG was added when the culture reached the mid-exponential growth phase at an OD₆₀₀ of approximately 0.8. The bacterial cells were cultured at 37°C for 4h and collected. WaaA and WaaA* were purified via immobilized metal affinity chromatography on a nickel cephalosporin HP column. The proteins were desalted with PBS (pH 7.2) and 20% glycerol and concentrated using a protein concentration column (Mamat et al., 2009). Protein concentrations were determined using a BCA protein assay kit.

To compare the circular dichroism (CD) spectra of wild-type WaaA and WaaA*, a Jasco J-815 spectropolarimeter (Greenfield, 1999; Kelly et al., 2005) was used. Protein solutions (0.1–0.35 mg/ml) were quantified at 200–260 nm. Spectra were obtained with a scan speed of 50 nm/min and 0.5-nm data pitch. The percentage of secondary structure was estimated using online software connected to Jasco J-815.

Protein Data Bank (PDB) was used to conduct homology search (Rose et al., 2015). For WaaA, the structure of membrane-embedded monofunctional glycosyltransferase WaaA from Aquifex aeolicus (PDB code: 2XCI.A) was used as the template based on the quality of models produced. Multiple sequence alignment was performed using Clustal Omega from UniProt (Yip et al., 2008). The ESPript server was used to predict the secondary structure of WaaA (Robert and Gouet, 2014). The Alphafold v2.3.1 monomer_casp14 model was used to predict the structures of WaaA and WaaA* (Jumper et al., 2021). The biological and functional insights derived from the predicted models were verified by matching the models with the protein function database.

Yep-phi is routinely used as a diagnostic phage for identifying Y. pestis, and no natural Yep-phi phage-resistant Y. pestis isolate has been reported (Zhao et al., 2011). In this study, we exposed cultures of Y. pestis 614F to serially diluted phage Yep-phi. The surviving clones were selected and amplified continuously. Figure 1A illustrates how resistant mutants were identified by rounds of progressive challenges with different phage titers. Phage spotting assays revealed that after 12 rounds of phage challenge, the derivate of 614F (strain S12) showed decreased phage susceptibility (resistant to phage titer 10⁵ PFU), whereas later derivates (S38 and S56, with 38 and 56 rounds of challenges, respectively) were fully resistant to phage titer 10⁸ PFU ( Figure 1B ). The EOP assay showed a similar result; the EOP value for S12 was 3.2 × 10⁻⁶, whereas the EOP values for S38 and S56 were<1.68 × 10⁻⁹ ( Figure 1C ). Our results confirmed that the 614F strain gradually developed full resistance to phage Yep-phi under continuous stress. Residual PFU percentages were used to evaluate the adsorption capability of 614F and its derivates S12, S38, and S56. All derivate strains lost their ability to adsorb phages ( Figure 1D ), which was confirmed through the temporal kinetics of adsorption assays ( Figure 1E ).

To examine the genetic variations responsible for phage resistance, we sequenced the genomes of strain S56 and its ancestor strain 614F. A control strain S88 (56 passages without phage challenges) was sequenced to rule out unrelated mutations. We identified three mutations in S56 ( Supplementary Figure S1 ): a 9-bp in-frame deletion (₂₄₉GTCATCGTG₂₅₇) in waaA (gene ID: YPO0055), a 10-bp frame shift deletion (₁₅CCGGTGATAA₂₄) in cmk (gene ID: YPO1391), and a 1-bp frame shift deletion (A₅₃₈) in ail (gene ID: YPO2905), which are annotated according to Y. pestis CO92. The mutations were named waaA*, cmk*, and ail*, respectively, and mismatch-PCR was used to screen their presence in passage stocks. The findings revealed that waaA* occurred in passage 12 (strain S12), cmk* occurred in passage 38 (strain S38) in addition to waaA*, and ail* occurred in passage 56 (strain S56) in addition to waaA*/cmk*.

To determine whether a similar phage-resistant profile can be reproduced and the roles of mutations in waaA (9-bp deletions), cmk (10-bp deletions), and ail (1-bp deletion), we constructed single, double, or triple mutants derived from strain 201. Strain waaA* was partially resistant to phage Yep-phi (resistant to 3.15 × 10³ PFU, Figure 2A ), with an EOP value of 5.4 × 10⁻⁶ ( Figure 2B , phage titer 1.05 × 10¹⁰ PFU). Moreover, waaA*/cmk* and waaA*/cmk*/ail* were fully resistant to phages with EOP< 1.05 × 10⁻¹⁰, which is much lower than that for waaA* ( Figure 2B ). Adsorption assays indicated that all mutants had a significantly lower ability to adsorb phage Yep-phi, and the residual PFU percentages among the three mutants were indistinguishable ( Figures 2C, D ). These results suggest that the waaA* mutation affects the binding between Y. pestis and its phage.

Phage adsorption defects were found in strain waaA* and strain 614F carrying the waaA* mutation. The waaA* mutant and the waaA null mutant ΔwaaA displayed similarly decreased phage susceptibility (resistant to 4.35 × 10³ PFU phage), with decreased EOP ( Figure 3A ). In the complementation test, the phage susceptibility of waaA* supplemented with plasmid pACYC184_waaA was restored and comparable to that of wild-type strain 201 ( Figures 3A, B ). In the adsorption kinetics assays, the phage adsorption ability of waaA* was similar to that of ΔwaaA, and waaA* mutant complemented with plasmid pACYC184_waaA showed similar phage adsorption ability to the wild-type strain, confirming the anti-adsorption effect of the waaA* mutation ( Figures 3C, D ).

Because the waaA gene, encoding 3-deoxy-D-mannooctulosonic acid (KDO) transferase, is involved in LPS synthesis (Chung and Raetz, 2010), we hypothesize that the mutation in waaA* affects the adsorption of Yep-phi phage because of failure to synthesize LPS, which is the phage adsorption receptor. At 26°C, Y. pestis expresses rough LPS, which does not contain the O-polysaccharide that is considered the phage receptor (Kiljunen et al., 2011). Rough LPS was expressed in 201 WT but was not stained on the gel by silver (a characteristic of a core-lacking LPS) in either ΔwaaA or waaA* ( Figure 3E ), which confirmed the failure of complete LPS core synthesis in the two mutants (Meredith et al., 2006; Dentovskaya et al., 2011). CD spectra assays demonstrated that wild-type WaaA has two negative absorption bands at 209 and 222 nm, which are characteristic of α-helical proteins, and the CD spectrum of WaaA* protein is slightly different from that of wild-type strain ( Figure 3F ). Meanwhile, the build-in software of the spectropolarimeter predicted decreased α-helix and increased β-sheet in WaaA* protein comparing with the wild-type WaaA ( Figure 3G ).

The 9-bp (₂₄₉GTCATCGTG₂₅₇) in-frame deletion in waaA resulted in the ₈₄TMT₈₆ deletion in WaaA. To further determine the consequence of the ₈₄TMT₈₆ deletion in WaaA, we compared the predicted structures of wild-type WaaA and mutated WaaA*. Sequence alignment of WaaA and WaaA* showed the presence of charged residue matching the monofunctional glycosyltransferase WaaA of A. aeolicus (PDB code: 2XCI.A) (Schmidt et al., 2012; Rose et al., 2015). In A. aeolicus, residues S54 and R56 of WaaA had been proven vital for its KDO transferase activity, and these two residues together with S28 and E31 provide necessary hydrogen bonds to bind tetraacyl-4'-phosphate lipid A during KOD transferring process as part of the receptor-substance binding site of WaaA (Schmidt et al., 2012). Secondary structure predictions of WaaA and WaaA* indicated that the ₈₄TMT₈₆ deletion occurred in the junction of β-sheet 2 (β2) and α-helix3 (α3), right next to residues S54 and R56 of A. aeolicus WaaA ( Supplementary Figure S2 ). Sequence alignment of multiple bacterial WaaA showed that the ₈₄TMT₈₆ region is conserved among multiple bacterial species ( Supplementary Figure S3 ). Additionally, Alphafold analysis of WaaA and WaaA* proteins demonstrated that their predicted structures are slightly different from each other ( Supplementary Figures S4A–C ). The ₈₄TMT₈₆ is located at an α helix-loop-β turn connection in WaaA protein, and the deletion of ₈₄TMT₈₆ of resulted in a shorter length of the α helix, followed by a shift of approximately 2.7 Å in the loop and a 6.8° deflection of the in the β turn ( Supplementary Figure S4D ).

Our results suggest that the ₈₄TMT₈₆ deletion in WaaA hinders its binding to the precursor of lipid A, and causes the loss of its glycosyltransferase function. The three amino acids in-frame deletion in waaA* resulted in a core-lacking LPS of Y. pestis, leading to phage resistance similar to ΔwaaA.

cmk encodes a cytidylate kinase that catalyzes phosphoryl transfer from ATP to (d)CMP (Tsao et al., 2015). Mutant waaA*/cmk* was more resistant to phage lysis than waaA*. However, the role of the cmk mutation against phage lysis remains unknown. To address this, we constructed cmk*, which contains a frameshift mutation (₁₅CCGGTGATAA₂₄ deletion). We performed infection assays using cmk*, C_cmk* (a complementary strain), and Δcmk (cmk null mutant) simultaneously and found that cmk* and Δcmk were resistant to a high phage titer (3.15 × 10⁷ PFU) ( Figure 4A ), with EOP values of 5.8 × 10⁻⁵ and 4.5 × 10⁻⁵, respectively ( Figure 4B ). C_cmk* restored susceptibility, comparable with wild-type 201. Notably, cmk* and Δcmk have similar adsorption capacity compared with WT 201 ( Figures 4C, D ), suggesting that cmk*-mediated phage resistance is irrelevant for the adsorption of Y. pestis and phage particles. The 10-bp (₁₅CCGGTGATAA₂₄) deletion of cmk led to premature termination of Cmk translation. The mechanism of Cmk-related phage resistance needs further investigation.

In Y. pestis, Ail is an outer membrane protein that contains 182 amino acids. Our previous study revealed that Ail interacts with the phage tail fiber protein and contributes to phage adsorption in Y. pestis (Zhao et al., 2013). In this study, we identified the A₅₃₈ deletion in ail in strain S56 (ail*, Table 1 ), which is at the very end of the ail coding region. ail* or Δail was as susceptible to phages as wild-type 201, with comparable EOP values ( Figures 5A, B ). Moreover, the ail* mutant showed similar phage adsorption defects as the null mutant Δail ( Figures 5C, D ). The addition of waaA* or cmk* to ail* did not make a difference in phage adsorption ( Figures 5C, D ); however, it significantly increased resistance to phage compared with the ail* null strain ( Figures 4B , 5B ). Our results indicate that ail* (1-bp deletion at A₅₃₈) exhibited phage adsorption resistance, consistent with the effects of disrupting phage adsorption in Δail.

The waaA*/cmk* mutant is resistant to high-titer phage attacks ( Figures 2A, B ), and the ail* mutant is somewhat sensitive to phages even though it adsorbs phages less effectively. We wondered whether the evolution of phage resistance due to the waaA*/cmk* mutations was a liability that was compensated by the ail mutation or why the weakly lysis-resistant ail* mutation occurred in Y. pestis when the waaA*/cmk* mutation already renders the strain completely phage resistance. The 614F derivate S12 grows slower in LB than 614F. The growth of S38 is even slower than that of S12, which suggests the great fitness costs accompanying the phage resistance phenotype of these two derivates. Interestingly, S56 grew much better than S38, and its growth recovered to the level of 614F ( Figure 6A and Supplementary Figure S5A ). Because waaA*/cmk*/ail* mutations occurred sequentially in S12, S38, and S56, we assumed that waaA* and cmk* mutations influence the growth of the mutants in addition to conferring phage resistance onto them. Growth curves showed a deceleration of bacterial growth for cmk*, especially for waaA*, while the subsequent ail* mutation restored the mutant growth, as measured by calculating area under the curve ( Figures 6B, C ; Supplementary Figures S5B, C ). Our results indicate that the growth deficiency in waaA*/cmk* mutants can be compensated by the A₅₃₈ deletion of ail.