Discovery and biosynthesis of a novel diaminopropane-modified thymine in phage DNA

Abstract

Bacteriophage genomes exhibit exceptional diversity in nucleotide modifications, which primarily function to counteract host defense systems. However, the diversity of phage DNA hypermodifications remains largely unexplored in post genome era. Here, we discovered a novel thymine hypermodification, α-1,3-diaminopropanylthymine (α-dapT), in the Acinetobacter baumannii phage SH-Ab 15599, and elucidated its biosynthetic pathway, featuring the phage-encoded key diamine DNA transferase (DADT, formerly αGPT-Pplase2). DADT utilizes the metabolite 1,3-diaminopropane, which is the major polyamine in the host to modify phage DNA. DADT exhibits broad in vitro substrate specificity but a strong in vivo preference for 1,3-diaminopropane. Structural and mutagenesis analyses revealed the molecular basis for substrate recognition and catalysis. The α-dapT modification occurs preferentially at TG dinucleotides and confers resistance to multiple host restriction enzymes. Furthermore, RNA-seq analysis showed that phage infection upregulates genes for 1,3-diaminopropane synthesis, and downregulates genes for 1,3-diaminoproprane consumption to supply the modification precursor. Given that 1,3-diaminopropane functions as a key regulator in mobility of A. baumannii, its metabolic reprograming may impair host biofilm formation.

Graphical Abstract

Graphical Abstract.

Introduction

Beyond the four canonical DNA bases, exceeding 40 distinct DNA base modifications have been reported in nature [1], and bacteriophage genomes harbor exceptionally rich repertoires of nucleobase modifications, exhibiting substantial diversity in biosynthetic pathways [2]. Non-canonical DNA base modifications in bacteriophages function primarily to evade host-encoded antiviral barriers, particularly restriction-modification and CRISPR–Cas systems [3–8].

In recent years, an expanding spectrum of bacteriophage DNA modifications has been identified. These efforts have not only identified previously unrecognized modified bases, but also elucidated biosynthetic pathways responsible for installing both newly discovered and long-known modifications [2, 4, 9–12], highlighting rapid progress in phage DNA modification systems. In 2018, two previously unrecognized thymine hypermodifications, 5-NeOmdU in Salmonella phage ViI and 5-NedU in Pseudomonas phage M6, were discovered [11]. In 2021, Lee and colleagues elucidated the biosynthetic pathway of these thymine hypermodifications [12]. The post-replication modification of the pathway initiates from 5hmdU-DNA, which is first phosphorylated by 5-HMUDK to generate 5-phosphomethyl-deoxyuridine (5-PmdU) (Supplementary Fig. S1, Step I). Then, distinct phages employ their αGPT-Pplase2 (hereafter referred to as Pplase2) enzymes together with specific ligands to generate hypermodified DNA (Supplementary Fig. S1, Step II). These modified nucleosides can either exist in the genome as an end product or can be further diversified by one or more downstream tailoring enzymes to yield the final hypermodified bases (Supplementary Fig. S1, Step III). The repertoire of tailoring enzymes includes radical SAM isomerases, PLP-dependent decarboxylases, flavin-dependent lyases, and acetyltransferases, collectively expanding the chemical diversity of thymine hypermodifications. This framework also led to the identification of an additional modification, 5-AcNmdU, in Pseudomonas phage PaMx11 [12].

Intriguingly, thymine hypermodifications catalyzed by Pplase2 of phages exhibit remarkable diversity in both ligand utilization and chemical bond formation. For example, the Pplase2 of Delftia phage PhiW-14 transfers putrescine to the 5-methyl group of thymine, forming an amide (N-C) bond [13]. And the Pplase2 of phage M6 employs glycine to form the intermediate modification α-GlyT, where the substituent is attached to the thymine 5-methyl group also through an amide bond [11]. However, the Pplase2 from phage ViI uses serine to generate 5-NeOmdU, in which the installed moiety is connected to the 5-methyl group of thymine via an ester (O-C) bond [11]. The remarkable chemical diversity of thymine modifications suggests that the enzymatic repertoires and reactions that underpin these pathways could be substantially broader than currently appreciated.

Materials and methods

Experimental materials

Information on the manufacturers and catalog numbers of the small-molecule compounds is provided in Supplementary Table S1. All primers used for cloning, as well as the 5′-biotinylated primer used for generating DNA substrates for enzymatic activity assays, were synthesized by Sangon Biotech (Shanghai, China). dNTPs and NTPs were purchased from Yeasen (Cat# 10122ES74, 10133ES03).

Experimental procedures

Bacteria strain and phage propagation

Acinetobacter phage SH-Ab 15599 and its host strain were kindly provided by Prof. Ping He. The host was cultured in LB medium at 37°C with shaking at 200 rpm. Phages were amplified using the double-layer agar method as previously described [14]. Briefly, phages were propagated on the host bacteria at 37°C overnight. Plates were flooded with SM buffer [50 mM Tris–HCl, pH 7.5, 100 mM NaCl, 8 mM MgSO₄, and 0.01% (w/v) gelatin] and shaken at 80 rpm for 4 h. The eluate was collected, supplemented with chloroform to 2% (v/v), incubated at 4°C for 30 min, and centrifuged at 6500 rpm for 15 min at 4°C. The supernatant was treated with DNase I and RNase A (1 μg/ml each) for 30 min at room temperature. NaCl was then added to a final concentration of 0.5 M, and the mixture was incubated on ice for 1 h, followed by centrifugation at 11 000 × g for 10 min at 4°C. Polyethylene glycol (PEG) 8000 was added to a final concentration of 10% (w/v), mixed thoroughly, and incubated at 4°C overnight. Phage particles were pelleted by centrifugation at 8500 rpm for 20 min at 4°C, resuspended in SM buffer, extracted with an equal volume of chloroform, and centrifuged at 4000 × g for 10 min. The aqueous phase was collected as the concentrated phage stock. Phage titers were determined by plaque assay and expressed as PFU/ml.

Genomic DNA extraction and purification of bacteriophage 15599

Genomic DNA of phage 15599 was extracted using a modified version of a previously described method by Sambrook and Russell [15]. The phage supernatant was treated with DNase I and RNase A at final concentrations of 20 and 60 μg/ml, respectively. After gentle mixing and incubation at 37°C for 30 min, proteinase K and sodium dodecyl sulfate (SDS) were added to final concentrations of 50 μg/ml and 0.5% (w/v), respectively, followed by incubation at 56°C for 1 h. After cooling to room temperature, an equal volume of phenol–chloroform–isopentyl alcohol (25:24:1; Magen, C493) was added. The mixture was gently inverted and centrifuged at 5000 rpm for 5 min at room temperature. The upper aqueous phase was transferred to a new tube, and the extraction was repeated twice. Sodium acetate (pH 7.0) was then added to a final concentration of 0.3 M, followed by two volumes of absolute ethanol. After incubation at room temperature for 30 min, DNA was pelleted by centrifugation at 12 000 rpm for 10 min at 4°C. The pellet was washed twice with 70% ethanol, air-dried, and resuspended in DNase/RNase-free deionized water (Share-bio, SB-DW001). DNA quality was assessed by agarose gel electrophoresis. If residual RNA was detected, DNA was further purified using DNA selection Beads (Yeasen, 12601ES08). Briefly, 0.45 volumes of beads were added to the DNA solution and incubated at room temperature for 5 min. The beads were captured on a magnetic rack, washed twice with 80% ethanol, air-dried, and the purified genomic DNA was eluted in DNase/RNase-free deionized water for subsequent experiments.

Gene synthesis and cloning

Codon-optimized gene fragments encoding 15599 5-HMUDK, Pplase2, dCMP deaminase (DctD), and dTTPase/dUTPase, as well as YuA 5-HMUDK and Pplase2, were synthesized by Genewiz Inc. (Suzhou, China) or Sangon Biotech (Shanghai, China). The 15599 dUMP hydroxymethylase (dUMP HMase) gene was amplified from phage genomic DNA by polymerase chain reaction (PCR).

For crystallization and enzymatic assays, the genes encoding 15599 5-HMUDK and YuA 5-HMUDK were cloned into the pET-28a(+) to generate constructs with an N-terminal His₆ tag. The 15599 Pplase2 and YuA Pplase2 genes were subcloned into pET-28a(+)-SUMO, yielding constructs with an N-terminal His₆-SUMO tag. The 15599 dTTPase/dUTPase gene was inserted into pRSFDuet-1 with an N-terminal MBP tag followed by an HRV 3C protease cleavage site.

For co-expression of 15599 genes involved in thymine hypermodification biosynthesis in Escherichia coli, the dUMP HMase and dTTPase/dUTPase genes were cloned into the multiple cloning sites (MCS I and MCS II) of pETDuet-1, respectively, generating pETDuet-HMase-dTTPase/dUTPase. The DctD and 5-HMUDK genes were cloned into MCS I and MCS II, respectively, of a laboratory-constructed pETDuet-SUMO-Apr vector carrying apramycin resistance, generating pETDuet-SUMO-Apr-DctD-5-HMUDK.

Point mutations in 15599 Pplase2 were introduced by site-directed mutagenesis and confirmed by DNA sequencing. Mutant proteins were expressed and purified using the same procedures as for the wild-type proteins.

Expression and purification of enzymes

Constructed plasmids were transformed into E. coli ArcticExpress (DE3) pRARE2 cells (Weidi, EC2021). Transformants were cultured in 2× YT Broth (Sangon Biotech, A507019) supplemented with 50 μg/ml kanamycin, 34 μg/ml chloramphenicol, and 20 μg/ml gentamicin at 37°C to an OD₆₀₀ of 0.6–0.8. Protein expression was induced with 250 μM isopropyl-β-D-thiogalactopyranoside (IPTG) at 16°C for 18–20 h. Cells were harvested by centrifugation at 6000 rpm for 10 min at 4°C.

Purification buffers were as follows: buffer A (20 mM HEPES, pH 7.5, and 300 mM NaCl) for 15599 5-HMUDK, YuA 5-HMUDK, and YuA Pplase2; and buffer B [50 mM HEPES, pH 7.5, 500 mM NaCl, and 5% (v/v) glycerol] for 15599 Pplase2 and dTTPase/dUTPase. Cell pellets were resuspended in the corresponding buffer supplemented with 0.2 mg/ml lysozyme, 0.03% Triton X-100, 0.02 mg/ml DNase I, and 1 mM phenylmethylsulfonyl fluoride. Suspensions were flash-frozen in liquid nitrogen and stored at −80°C. For purification, samples were thawed and incubated at 25°C for 30 min to facilitate lysis. Streptomycin sulfate was added to a final concentration of 1.1% (w/v), and lysates were centrifuged at 18 000 rpm for 40 min at 4°C. The supernatant was collected for subsequent purification.

His₆-tagged and His₆-SUMO-tagged proteins were purified by Ni-NTA affinity chromatography (Qiagen, 30430) using columns equilibrated with buffer A or B. After washing with 10–20 column volumes (CV) of the corresponding buffer containing 20 mM imidazole, the bound proteins, including His₆-tagged 15599 5-HMUDK and YuA 5-HMUDK, and His₆-SUMO-tagged 15599 Pplase2, were eluted with 5–10 CV of buffer containing 200 mM imidazole. The eluted proteins were dialyzed overnight against 2 l buffer A or B to remove imidazole. His₆-SUMO-tagged YuA Pplase2 was incubated with ULP1 protease at 4°C overnight to remove the His₆-SUMO tag. The tag-free YuA Pplase2 was subsequnetly collected with 5–10 CV of buffer A containing 20 mM imidazole. YuA Pplase2 used for crystallization was further purified by size-exclusion chromatography on a Superdex 200 Increase 10/300 column (Cytiva, 28990944) pre-equilibrated with buffer C (20 mM HEPES, pH 7.5, and 200 mM NaCl). MBP-tagged 15599 dTTPase/dUTPase was purified using Amylose Resin (NEB, E8021L) pre-equilibrated with buffer B. After washing with 10 CV of buffer B, the protein was eluted with buffer B supplemented with 20 mM maltose. All proteins were concentrated using centrifugal concentrators (Millipore) and analyzed by 4%–20% sodium dodecyl sulphate–polyacrylamide gel electrophoresis (SurePAGE, GenScript).

Crystallization, data collection, and structure determination

Crystallization was performed using the hanging drop vapor diffusion method at 20°C. Purified YuA Pplase2 was concentrated to 25 mg/ml in buffer C. For complex formation, YuA Pplase2 (20 mg/ml) was preincubated with synthesized 5hmdU-DNA substrates and glycine at a molar ratio of 1:1.2:5. Crystals were obtained by mixing the protein solution with reservoir solution containing 0.1% (w/v) n-octyl-β-D-glucoside, 0.1 M sodium citrate, pH 5.5, and 22% (w/v) PEG 3350. Crystals were cryoprotected in reservoir solution supplemented with 30% ethylene glycol and flash-cooled into liquid nitrogen.

Diffraction data were collected at BL18U1 of National Facility for Protein Science Shanghai, and processed with HKL3000 [16]. Data collection statistics are summarized in Supplementary Table S2. The structure was determined by molecular replacement using the AlphaFold2-predicted structure of YuA Pplase2 as the search model. Model building was performed with Coot [17], and refinement was carried out using Phenix [18]. Structure figures were generated using ChimeraX [19]. Refinement statistics and model quality metrics are provided in Supplementary Table S2.

Preparation of 5hmdU-DNA substrates

For co-crystallization of YuA Pplase2 and 5hmdU-DNA, a 13-bp DNA duplex was synthesized (WuXi AppTec) and annealed from single-stranded oligonucleotides: forward strand, 5′-CGAT(^5hmU)GCCCATCG-3′; reverse-strand, 5′-CGAT(^5hmU)GGCCATCG-3′.

The 500-bp 5hmdU-DNA substrate used for enzyme activity assays was generated by PCR amplification of a randomly selected region of 15599 genomic DNA using DreamTaq DNA polymerase (Thermo Scientific, EP0703). In the PCR reaction, 5hmdUTP (Trilink, N-2059-10) was used in place of dTTP, together with dATP, dGTP, and dCTP. The forward primer was 5′-biotinylated to enable affinity purification of DNA products after enzymatic reactions.

In vitro DNA modification assay

DNA modification assays were performed at 37°C using a modified protocol described previously [12]. Reaction mixture (50 μl) contained 4 μg of 5-HMUDK, 4 μg Pplase2, 500 ng biotinylated 500-bp 5hmdU-DNA, and 5 mM ligand in reaction buffer [25 mM Tris–HCl, pH 7.5, 25 mM KCl, 1 mM ATP, 5 mM MgCl₂, and 5 mM Tris(2-carboxyethyl)phosphine]. After incubation for 2 h, reactions were quenched by adding an equal volume of quenching/bead-binding buffer [20 mM Tris–HCl, pH 8.0, 2 M NaCl, 2 mM ethylenediaminetetraacetic acid (EDTA), and 5% (w/v) polyvinyl alcohol (average MW 30 000–70 000, Sigm–Aldrich, P8136)]. Streptavidin magnetic beads (NEB, S1420S), pre-equilibrated with bead-washing buffer (25 mM Tris–HCl, pH 7.5, 500 mM NaCl, and 1 mM EDTA), were added to the quenched reactions, gently mixed, and incubated at room temperature for 10–15 min. DNA-bound beads were captured using a magnetic rack and the supernatant was discarded. The beads were washed twice with bead-washing buffer and resuspended in 43 μl ddH₂O. Subsequently, 5 μl nucleoside digestion buffer and 2 μl nucleoside digestion mix (NEB, M0649S) were added, and the mixture was incubated at 37°C for 12 h. After digestion, the beads were removed using a magnetic rack, and the supernatant was collected for liquid chromatography-mass spectrometry (LC-MS) analysis.

To investigate the metal-ion dependence of Pplase2-catalyzed reaction, a two-step enzyme activity assay was performed. First, 5-HMUDK was used to phosphorylate 5hmdU in the 5hmdU-DNA substrate in the presence of 1 mM MgCl₂. After 1 h, EDTA was added to the experimental group to a final concentration of 2 mM to chelate metal ions, whereas the control group received no EDTA. Both reactions were incubated at room temperature for 30 min. Pplase2 was then added to initiate the second catalytic step. Reactions were performed either without additional metal ions or with supplementation with 5 mM MgCl₂ or other divalent metal ions. After 2 h, reactions were quenched, and DNA products were purified, enzymatically hydrolyzed to deoxyribonucleosides, and analyzed by LC-MS as described above.

LC-MS analysis

The digested samples were analyzed using a Shimadzu UHPLC LC-30A system coupled to an AB SCIEX Triple 4600 spectrometer (QTOF). Chromatographic separation was performed on a Hypersil GOLD aQ column (1.9 μm particle size, 150 mm × 2.1 mm, Thermo Scientific) using gradient elution. Mobile phage A consisted of water containing 1% acetontrinile and 0.1% formate and mobile phase B consisted of acetontrinile. The gradient was programmed from 100% A to 0% A, followed by re-equilibration to initial conditions. The flow rate was 0.6 ml/min, and the column temperature was maintained at 25°C. Detection was performed using an electrospray ionization source in positive ion mode. Product formation [α-1,3-diaminopropanylthymine (α-dapT)] and residual substrate (5hmdU) were quantified by peak integration. The enzymatic activity of 15599 Pplase2 was calculated as the ratio of α-dapT to the sum of α-dapT and remaining 5hmdU. All experiments were performed in triplicates and data are presented as the standard error of the mean. Mass spectrometric data processing and chromatogram generation were performed using the MestReNova, and graphs were generated using Origin and GraphPad Prism.

Colorimetric pyrophosphate assay for dTTPase/dUTPase

The enzymatic activity of dTTPase/dUTPase was measured at 37°C using a previously described method [20]. Reaction (200 μl) contained 0.2 μM dTTPase/dUTPase, 0.5 mM substrate nucleotide, and 0.3 μM inorganic pyrophosphatase 1 in reaction buffer consisting of 50 mM HEPES, pH 7.5, 500 mM NaCl, 1 mM Co²⁺, and 1 mM Tris(2-carboxyethyl)phosphine. The tested substrates included dUTP, dTTP, 5hmdUTP, other dNTPs, and NTPs. After incubation for 2 h, reactions were terminated by heating at 95°C for 10 min. An aliquot of the reaction mixture (105 μl) was transferred to a 96-well plate, mixed with an equal volume of freshly prepared colorimetric reagent, shaken, and incubated at room temperature for 20 min. Absorbance was measured at 660 nm using a FlexStation 3 microplate reader.

The colorimetric reagent was prepared by adding 5 ml solution A (10% ammonium molybdate in 10 N H₂SO₄) to 40 ml deionized water, followed by dissolution of 2.5 g FeSO₄·7H₂O. The final volume was adjusted to 50 ml with deionized water.

Chemical synthesis of α-dapT

α-dapT was synthesized by WuXi AppTec. Synthetic procedures and characterization data are provided in the Supplementary data. Structures were confirmed by nuclear magnetic resonance (NMR) and LC-MS/MS, and purity (≥99%) was determined by high performance liquid chromatography (HPLC).

Restriction enzymes digestions of 15599 genomic DNA

Approximately 500 ng of 15599 phage genomic DNA was digested with restriction enzymes in a 20-μl reaction according to the manufacturer’s instructions. All enzymes were purchased from New England Biolabs (NEB, USA).

Co-expression of genes in E.coli

To co-express 15599 genes involved in thymine hypermodification biosynthesis in E. coli, the constructed plasmids pETDuet-HMase-dTTPase/dUTPase, pETDuet-SUMO-Apr-DctD-5-HMUDK, and pET-28a(+)-SUMO-15599-Pplase2 were co-transformed into E. coli C43(DE3) pLysS cells (Weidi, EC1041). The control group was transformed with the corresponding empty vectors. Transformants were cultured in LB medium supplemented with 100 μg/ml ampicillin, 50 μg/ml kanamycin, and 50 μg/ml apramycin at 37°C to an OD₆₀₀ of 0.6–0.8, followed by induction with 200 μM IPTG at 16°C. For designated experimental groups, 1,3-diaminopropane was added to a final concentration of 5 mM. After induction for 18–20 h, 4–5 ml cultures were harvested by centrifugation at 4000 rpm for 10 min at 4°C. Bacterial genomic DNA was extracted using Bacterial DNA Kit (Yeasen, 18806ES50), and DNA quality was assessed by agarose gel electrophoresis. Genomic DNA (500 ng) was then digested with nucleoside digestion mix in a 50-μl reaction at 37°C for 12 h. The resulting products were centrifuged and subjected to LC-MS analysis.

Total RNA extraction and RNA sequencing

For RNA-seq analysis, host cells were grown at 37°C to an OD₆₀₀ of ∼0.6 and infected with phage 15599 at a multiplicity of infection (MOI) of 20 to ensure synchronous infection. Samples for RNA extraction were collected from uninfected cultures, used as negative controls, and from infected cultures at multiple time points post-infection: 0, 5, 10, 15, 20, 30, and 40 min. Three independent biological replicates were prepared for each condition. Total RNA was extracted using Bacterial RNA Kit (Yeasen, 19301ES50) according to the manufacturer’s protocol. RNA concentration was measured using a NanoDrop One Spectrophotometer (Thermo Fisher Scientific). The extracted total RNA was submitted to Shanghai Sangon Biotech for quality assessment, library preparation, and sequencing. Library quality was assessed using an Agilent 4200 Bioanalyzer. Ribosomal RNA was depleted using a bacterial rRNA depletion kit (Vazyme, NR407-2). The rRNA-depleted RNA samples were then used for library preparation with VAHTS Stranded mRNA-seq V2 Library Prep Kit for Illumina (Vazyme, NR601-01). Paired-end 150-bp (PE150) sequencing was performed on an Illumina NovaSeq X plus platform.

Determination of the optimal multiplicity of infection

The optimal MOI for phage 15599 was determined by infecting the host strain at MOIs of 10, 1, 0.1, 0.01, and 0.001. Phage suspensions and host cultures were mixed and incubated at 37°C with shaking at 180 rpm for 4 h. The cultures were then centrifuged at 5000 rpm for 10 min, and the supernatants were filtered to remove residual cells. Phage titers were determined by the double-layer agar plaque assay, and the MOI that produced the highest titer was defined as the optimal MOI. All experiments were performed in triplicate.

Crystal violet staining assay for biofilm formation

Biofilm biomass was quantified using a crystal violet staining assay with minor modifications to a previously reported method [21]. Briefly, 100 μl of host culture and 100 μl of phage 15599 suspension were added to each well of a 96-well plate at MOIs of 1.0, 0.1, 0.01, and 0.001. For the control group, 100 μl of host culture was mixed with 100 μl of LB medium. Plates were incubated at 37°C for 24 h to allow biofilm formation.

After incubation, planktonic bacteria were removed, and the wells were gently washed with phosphate buffered saline (PBS). The attached biofilms were fixed with 4% paraformaldehyde for 30 min and stained with 0.5% crystal violet solution for 10 min. Excess stain was removed by washing with PBS, and the plates were dried at 65°C for 1 h. Bound crystal violet was solubilized by adding 220 μl of 33% acetic acid to each well, and the absorbance was measured at 580 nm. The experiment was performed with 10 replicates.

Bioinformatic analyses

Pplase2 homolog identification and SSN construction

Homology searches were performed against an in-house sequence database using a locally installed version of jackHMMER program (HMMER v3.1b2 [22]). A total of 4445 homologous sequences were identified. Redundancy was reduced by clustering sequences at 90% identity and 90% coverage, yielding 2751 representative sequences for subsequent sequence similarity network (SSN) onstruction. SSNs were generated using EFI-EST (https://efi.igb.illinois.edu/efi-est/). Networks were visualized by Cytoscape 3.7.0 [23], using an E-value cutoff of 10⁻⁵⁰, corresponding to an approximate sequence identity of ≥30% between connected nodes.

Genomic context analysis

Genomic contexts were retrieved from coding sequence (CDS) records in NCBI genome entries (https://static.pubmed.gov/portal/portal.fcgi/). Each CDS was annotated against the Pfam-A database [24] using HMMER (HMMER 3.1b2) [22].

Multiple sequence alignment

Multiple sequence alignment (MSA) was generated using MAFFT (L-INS-i v7.505) [25]. Sequence probability logos highlighting conserved sites in the 15599, YuA, and ViI clusters were generated using Weblogo 3 [26].

Structural modeling and molecular docking

The DNA-Mg²⁺-bound complex of 15599 Pplase2 was modeled by integrating the crystal structure of DNA-Mg²⁺-glycine-bound YuA Pplase2 complex with the predicted structure using AlphaFold3 [27] (local installation, v3.0.0). Diamines of varying chain lengths were subsequently docked into the 15599 Pplase2-DNA-Mg²⁺ complex. Molecular docking was performed using Glide in extra-precision (XP) mode within the Schrödinger suite (Glide, Schrödinger, LLC, New York, NY, 2025) [28, 29]. For each docking run, 15 candidate binding poses were generated. The most plausible pose was selected based on docking score in combination with prior structural knowledge derived from the YuA Pplase2-DNA-Gly complex.

The dUTP-Mg²⁺-bound complex of 15599 dTTPase/dUTPase was predicted using AlphaFold3, and complexes with dCTP, dTTP, and 5hmdUTP (each with Mg²⁺) were generated by molecular docking based on this model. Molecular docking was performed as described above, and the final poses were selected based on consistency with the dUTP-bound conformation of 15599 dTTPase/dUTPase.

Protein pocket volumes of 15599 and PhiW-14 Pplase2 were calculated using the “trajectory_binding_site_volumes.py” module in the Schrödinger software (Desmond Molecular Dynamics System, D. E. Shaw Research, New York, NY, 2024. Maestro-Desmond Interoperability Tools, Schrödinger, New York, NY, 2025.) [30].

Calculation of strain energy

Conformational strain energy for each docked diamine was calculated as the difference between ligand-bound conformation energy and global minimum energy.

To identify the global minimum energy conformation, up to 64 conformers were generated for each diamine using ConfGenX [31] and ranked with the OPLS4 force field [29]. The conformer with the lowest OPLS4 energy was subsequently subjected to QM geometry optimization and single-point energy calculations. Ligand-bound conformation was prepared by performing constrained optimizations on the isolated ligand (with dihedral angles fixed). All quantum chemical calculations were performed using Gaussian 09 (revision E.01) [32]. Geometry optimizations were carried out at the B3LYP-D3/6–31G** level with the SMD solvation model, and single-point energy calculations at the M06-2X-D3/def2-TZVPP level with SMD. M06-2X [33] was selected for its reliable performance in describing noncovalent interactions in organic systems. The SMD [34] model was employed to account for solvation effects and to prevent electrostatic collapse that could lead to unrealistic geometries and overestimated strain energy values.

Identification of restriction endonucleases in Acinetobacter baumannii

A total of 68 putative restriction endonucleases (REases) were retrieved from the UniProt database by querying “restriction endonuclease” in A. baumannii. To identify bona fide REases, sequences were subjected to BLASTp [35] analysis against the Swiss-Prot database. Four candidates were identified based on sequence homology to characterized commercial REases: A0ABD5DFT8 (FPK87_26 630), A0A241ZGY5 (B9X95_05 225), A0AA44XXN2 (CV954_001 160) and A0A090B5P (G3N53_09 200).

Specifically, A0ABD5DFT8 shares 86.6% identity with P14633 (EcoRII), recognizing the CCWGG site. A0A241ZGY5 shows 59.9%, 44.3% and 32.7% identity with P33562 (BsuBI), P00640 (PstI) and D4ZX34 (AplI), respectively, all recognizing the CTGCAG site. A0AA44XXN2 shares 46.1% identity with Q04852 (CfrBI) recognizing the CCWWGG, and A0A090B5P9 shared 45.1% identity with P36433 (HpaII) recognizing the CCGG site (Supplementary Table S3). Based on these homology profiles, EcoRII, BsuBI, PstI, AplI, and CfrBI with recognition sites containing thymine recognize three distinct sequences. Therefore, PstI, BstNI (an isoschizomer of EcoRII), and StyI (an isoschizomer of CfrBI) were selected for genomic DNA digestion assays.

RNA-seq analyses

For RNA-seq analysis, three independent biological replicates were included for each condition (n = 3). Raw RNA-seq reads were processed for quality control and adapter trimming using fastp v0.24.0 [36]. Clean reads were aligned to the reference genomes of Acinetobacter phage SH-Ab 15599 (GenBank Accession: MH517022.1) and its host A. baumannii (GenBank Accession: JBXTSJ000000000) using HISAT2 v2.2.1 [37]. Alignment files were processed with SAMtools (v1.21) to calculate mapping coverage and sequencing depth. Gene expression levels were quantified using featureCounts v2.0.6 [38], with multimapping reads counted fractionally.

Downstream analyses were performed in the language R v4.4.2. Hierarchical clustering and heatmaps were generated via the pheatmap package (v1.0.12) based on Z-score normalized log₂(TPM + 1) values. Differential expression analysis was performed using DESeq2 (v1.46.0), with Log₂ fold changes (Log₂FC) for each post-infection time point calculated relative to the uninfected control (0 min). Statistical significance was evaluated using adjusted P-values (q-value). Genes with q-value <0.05 were considered significantly differentially expressed.

Statistical analysis

Statistical analyses for non-transcriptomic assays were performed using GraphPad Prism. Enzymatic activity assays were performed with three independent replicates and are presented as mean ± s.e.m. (n = 3). Biofilm biomass measurements and colony-forming unit (CFU) enumeration assays were performed with 10 independent replicates and are presented as mean ± s.e.m. (n = 10). Statistical significance was assessed by one-way analysis of variance (ANOVA). In the divalent metal ion supplementation assay, comparisons were made either with the untreated CK group or with the EDTA-treated group without additional metal ion supplementation, as specified in the corresponding figure legend. For biofilm biomass and viable bacterial counts after phage treatment at different MOIs, each treatment group was compared with the untreated control group. For non-transcriptomic assays, differences were considered statistically significant at P < 0.05; significance levels are indicated in the figures, with ^**** or ^#### denoting P < 0.0001.

Results

SSN analysis highlights a putative divergent Pplase2 from phage 15599

To investigate whether Pplase2 enzymes can utilize additional small-molecule ligands beyond glycine, serine, and putrescine to hypermodify 5hmdU-DNA in bacteriophages, we constructed an SSN comprising thousands of members of the Pplase2 family, displayed at BLAST E-value cutoff 10⁻⁵⁰ (Fig. 1A). In the SSN, proteins with known functions are color-coded. Many clusters remain functionally uncharacterized, underscoring the substantial diversity of the Pplase2 family. Notably, even clusters containing functionally characterized enzymes may still harbor undiscovered activities.

Figure 1.

Acinetobacter phage SH-Ab 15599 Pplase2 catalyzes the conversion of 5-hydroxymethyluridine to dapT. (A) SSN of Pplase2 proteins. The full SSN is shown on the left, and an enlarged local network containing functionally characterized Pplase2 clusters is shown on the lower right. Each node represents one Pplase2 protein. Proteins with known functions are indicated by colored circles, and 15599 Pplase2 is highlighted in orange. Chemical structures of known thymine hypermodifications are shown with the corresponding phages in which they occur. (B) AlphaFold3-predicted structural comparison of the substrate-binding pockets of Pplase2 from 15599 and PhiW-14. Divergent pocket residues are underlined, with residue numbering based on 15599 Pplase2. (C) HPLC analysis of nucleosides generated from enzymatic assays of 15599 Pplase2 using 5hmdU-DNA in the presence of Group 3 ligands. Mock-treated 5hmdU-DNA is shown as a control. A new peak corresponding to an unknown modified nucleoside is highlighted in orange. (D) LC-MS analysis of the new product peak generated by 15599 Pplase2. The fragment ion at m/z 199 corresponds to the nucleobase formed by neutral loss of the deoxyribose moiety (116 Da) from the precursor 2′-deoxynucleoside ion (m/z 315). (E) Chemical formula of α-dapT nucleoside deduced from high-resolution mass spectrometry, shown together with theoretical m/z values of candidate ligand-derived modified nucleosides. (F) Extracted ion chromatograms (EICs) at m/z 315.16 ± 10 ppm of 5hmdU-DNA treated with 15599 Pplase2 in the presence of 1,3-diaminopropane. Mock-treated 5hmdU-DNA and synthetic dapT are shown for comparison. (G) MS/MS spectra of dapT generated from 15599 Pplase2 treated 5hmdU-DNA and chemically synthesized dapT nucleoside.

We noted that the Pplase2 from Acinetobacter phage SH-Ab 15599 (hereafter 15599), a locally isolated phage available in our laboratory, resides in the same SSN cluster as the Pplase2 from Delftia phage PhiW-14 (Fig. 1A). However, the two enzymes are not directly connected and share only ∼35% sequence identity. Notably, putrescine—the small-molecule substrate used by the PhiW-14 Pplase2—is nearly absent in the host of phage 15599, A. baumannii [39, 40], raising the possibility that the 15599 Pplase2 catalyzes a distinct and previously unrecognized activity.

15599 Pplase2 catalyzes the formation of a novel hypermodification-dapT

Structural comparison of Pplase2 from phages 15599 and PhiW-14 revealed highly similar active sites that differ at only two positions, M18/I19 and A88/E88 (Fig. 1B). To identify the substrate of 15599 Pplase2, we compiled 53 candidate metabolites containing at least one amino group, shared by the known substrates of this family. These metabolites were divided into three pools according to their chemical properties (Supplementary Table S1) and screened in vitro with 5hmdU-DNA in the presence of purified 5-HMUDK and Pplase2 of 15599 (Supplementary Fig. S2). A new UV absorption peak was detected only with the third metabolite pool (Fig. 1C). LC-MS analysis assigned this peak and its fragment ion m/z values of 315.1675 and 199.1204, respectively, yielding a mass difference of 116.0471 and suggesting the generation of a previously uncharacterized modified nucleoside (Fig. 1D).

The theoretical m/z of the 1,3-diaminopropane-derived product (315.1668) differed from the measured value by only 1.9 ppm (Supplementary Table S1), identifying 1,3-diaminopropane as the substrate of 15599 Pplase2. By analogy to α-putrescinylthymine (putT), we proposed that the 15599 enzyme generates α-dapT (Fig. 1E).

To validate that 15599 Pplase2 catalyzes the formation of the proposed modification, we repeated the in vitro enzymatic assay using 1,3-diaminopropane as the sole ligand, and used a chemically synthesized dapT nucleoside (Supplementary Fig. S3) as the standard in LC-MS/MS analyses. The results showed that the modified nucleoside generated by 15599 Pplase2 had the same retention time and m/z as the synthetic standard (Fig. 1F and G), confirming its identity as dapT.

15599 Pplase2 preferentially transfers 1,3-diaminopropane within a defined diamine substrate range

To assess substrate selectivity, we examined 15599 Pplase2 activity in vitro using diamines of progressively increasing chain length. The enzyme accepted a broad but limited range of linear diamines, catalyzing transfer of 1,2-diaminoethane (dae), 1,3-diaminopropane (dap), 1,4-diaminobutane (putrescine, put), 1,5-diaminopentane (cadaverine, cad), and 1,6-diaminohexane (dahx), yielding the corresponding modified nucleosides daeT, dapT, putT, cadT, and dahxT, respectively. In contrast, no product was detected with 1,7-diaminoheptane (dahp), indicating an upper steric limit for amine transfer to the C5 hydroxymethyl group of 5hmdU (Fig. 2A and B). Thus, 15599 Pplase2 accommodates linear diamines within a defined chain-length window.

Figure 2.

15599 Pplase2 is a diamine DNA transferase with a preference for 1,3-diaminopropane. (A) HPLC traces of nucleosides from 5hmdU-DNA treated with 15599 Pplase2 in the presence of diamines with different chain lengths. Newly formed modified nucleosides are highlighted in different colors. dae, 1,2-diaminoethane; dap, 1,3-diaminopropane; put, 1,4-diaminobutane (putrescine); cad, 1,5-diaminopentane (cadaverine); dahx, 1,6-diaminohexane; dahp, 1,7-diaminoheptane. (B) MS/MS spectra of modified nucleosides generated by 15599 Pplase2 with the diamines shown in panel (A). HPLC analysis (C) and EICs (D) of 15599 Pplase2 activity on 5hmdU-DNA in the presence of mixed ligands at equal molar ratios.

We next examined whether 15599 Pplase2 could exploit amine donors beyond tested diamines. Among four common amines tested in vitro—histamine, tyramine, spermine, and spermidine—only histamine supported modification of 5-PmdU-DNA, generating a distinct nucleoside species not observed before (Supplementary Fig. S4). In addition, small monoamine screening revealed only marginal activity with ethylamine and n-propylamine, each producing trace levels of modified thymine (Supplementary Fig. S4). Together, these results indicate that 15599 Pplase2 is highly selective for diamines, therefore, we designated this enzyme as a diamine DNA transferase (DADT).

Despite its apparent substrate promiscuity, the enzyme exhibited a clear intrinsic preference for 1,3-diaminopropane under competitive conditions, with dapT emerging as the predominant product when the five productive diamines were provided at equal molar concentrations (Fig. 2C and D). Collectively, these findings demonstrate that 15599 Pplase2 functions as a DADT with a defined substrate range and a pronounced catalytic preference for 1,3-diaminopropane.

A Pplase2-DNA-ligand structure captures substrate recognition

To elucidate how Pplase2 enzymes recognize DNA and small-molecule substrates, we first sought to determine the crystal structures of 15599 Pplase2 and PhiW-14 Pplase2. However, repeated attempts to crystallize either protein, alone or as a SUMO fusion, were unsuccessful, likely owing to poor stability and limited crystallizability. We therefore turned to Pplase2 from Pseudomonas phage YuA, which shares 97% sequence identity with M6 Pplase2 (Fig. 1A). Consistent with this close relationship, we confirmed that YuA Pplase2, like M6 Pplase2, uses glycine as the small-molecule donor (Supplementary Figs S2 and  S5A and B) [12].

We next determined the crystal structure of YuA Pplase2 bound to 5hmdU-DNA and glycine at 2.2 Å resolution (Supplementary Table S2). The asymmetric unit contains one protein monomer bound to one single-stranded DNA molecule, and crystal packing generates an assembly in which two Pplase2 molecules engage one duplex DNA substrate (Supplementary Fig. S5C and D). In this complex, the target 5hmdU base is flipped from the duplex and inserted into the active site (Supplementary Fig. S5D–F). While this work was in progress, the structure of PaMx11 Pplase2 bound to unmodified DNA and glycine was reported [41]. Overall, the two complexes are highly similar (RMSD 1.07 Å over 265 Cα atoms). Notably, however, the glycine-binding mode differs substantially between them, potentially reflecting differences in the DNA substrates used for structure determination: in the YuA complex, the flipped-out nucleotide is a modified 5hmdU located in the middle of the duplex, whereas in the PaMx11 structure, the flipped-out nucleotide in an unmodified dT located in the first position of the 5′ strand. In the YuA structure, the amino group of glycine forms a hydrogen bond with E96, whereas its carboxylate group forms salt bridges with K159 and K248. The interaction with K159 is further stabilized by D24, while that with K248 is reinforced by D262 and Y257 (Supplementary Fig. S5F and G). Together, D24, E96, K159, K248, Y257, and D262 define the small-molecule-binding pocket and provide a structural framework for analyzing substrate recognition across the Pplase2 family.

Pplase2 defines a metal-dependent HhH-containing fold

Guided by a modeled structural complex of 15599 Pplase2 with Mg²⁺, 5-PmdU-DNA, and 1,3-diaminopropane, generated based on the YuA structure, we next performed systematic mutagenesis to investigate its catalytic mechanism. The active site of 15599 Pplase2 retains key features of the YuA enzyme but also shows substantial divergence (Fig. 3A and B; and Supplementary Fig. S6A).

Figure 3.

The substrate-binding pocket of 15599 Pplase2 governs metal-dependent catalysis, diamine recognition, and reprogrammable substrate specificity. (A) Structural comparison of the substrate-binding pockets of 15599 Pplase2 (AlphaFold3 model) and YuA Pplase2 (crystal structure). Residues that differ between the two pockets are underlined, and residues critical for ligand recognition are boxed. Residue numbering corresponds to 15599 Pplase2. (B) Schematic representation of the active site in the modeled complex of 15599 Pplase2 with Mg²⁺, 5-PmdU-DNA, and 1,3-diaminopropane. (C) Effects of mutations in residues involved in catalysis, metal binding, and ligand binding on 15599 Pplase2 activity. The heatmap indicates dapT formation. Data are from three independent replicates, n = 3. (D) Requirement of divalent metal ions for Pplase2 activity. Data are shown as mean ± s.e.m., n = 3. ^####P < .0001 versus CK; ^****P < .0001 versus (+)EDTA without additional supplementation of metal ions. (E) Sequence probability logos of key residues in 15599, YuA, and ViI Pplase2 clusters. (F) EICs of 5hmdU-DNA after reactions with wild-type or G227K/D238Y (GD/KY) mutant 15599 Pplase2 in the presence of dap or glycine, and with YuA Pplase2 in the presence of glycine. (G) MS/MS spectra of α-glyT generated from 5hmdU-DNA by the 15599 Pplase2 GD/KY mutant or by YuA Pplase2. (H) Summary of residues critical for Pplase2 activity across different clusters, together with representative phages and their ligand substrates. Residue numbering corresponds to 15599 Pplase2.

Mutation of the conserved catalytic cysteine in the Pplase2 superfamily abolished activity, confirming its essential role (Fig. 3C and Supplementary Fig. S6A). Previous profile–profile comparisons suggested that Pplase2 contains multiple helix–hairpin–helix (HhH) motifs and is related to the α-helical DNA glycosylase superfamily [42]. Although its overall topology displays modest similarity to several DNA glycosylases (TM-scores ∼0.5–0.6), neither CATH nor SCOP classifies the structure of YuA Pplase2 within any known structural family, indicating that it adopts a previously unrecognized fold. Notably, both the YuA and PaMx11 structures contain a metal ion in the active site, a feature not observed in characterized HhH DNA glycosylases.

To assess the functional importance of this metal, we used a two-step assay in which 5hmdU-DNA was first phosphorylated by 5-HMUDK and then treated with Pplase2 following EDTA treatment. EDTA chelation abolished dapT formation, whereas activity was restored by addition of Mg²⁺, Mn²⁺, Co²⁺, or Ca²⁺, but not Zn²⁺ (Fig. 3D). Consistent with this result, mutation of the metal-coordinating residues E222 and T223 eliminated activity (Fig. 3C and Supplementary Fig. S6A). These findings establish Pplase2 as a structurally distinct, metal-dependent HhH-containing family that catalyzes the transfer of amino-containing small molecules onto DNA to generate hypermodified bases, thereby distinguishing it functionally from canonical HhH DNA glycosylases, which recognize and excise damaged or inappropriate bases to initiate base excision repair.

A set of pocket residues specifies diamine recognition and donor selectivity in 15599 Pplase2

Mutational analysis of the 15599 Pplase2 substrate-binding pocket identified several residues critical for diamine transfer (Fig. 3C). Among these, D238E or D238Y abolished catalysis, whereas G227K reduced activity by ∼85%, and H230A lowered activity by 75%, indicating that these positions are important for utilization of 1,3-diaminopropane. D238 was particularly sensitive to substitution: beyond the dead mutations mentioned above, D238K caused an ∼80% decrease, whereas D238A retained substantial activity, highlighting the importance of the local chemical environment at this position. D243E and D243K reduced dapT production to ∼60% of wild-type levels. D238E/D243E double mutation markedly reduced activity, whereas D238K/D243K abolished it completely. Although D243A or D238A single mutation had little effect on activity, double mutation D238A/D243A greatly reduced activity, this suggests D238 and D243 act complementarily to create a negatively charged pocket required for diamine accommodation. A88E caused only a modest reduction in activity but substantially improved protein expression, which enabled interpretation of D25, whose single mutant failed to express. The near-complete loss of activity in the D25E/A88E double mutant therefore suggests that D25 also contributes importantly to catalysis, potentially by stabilizing the metal through electrostatic interaction. Additional substitutions had more moderate effects: Y54F and Y139F reduced activity to 40%–60% of wild-type levels, whereas H230S had only a modest effect (Fig. 3C).

Sequence and structural comparison further suggested that donor preference is encoded by cluster-specific pocket residues (Fig. 3E). In YuA Pplase2, the positions corresponding to G227 and D238 in 15599 Pplase2 are occupied by K248 and Y257, which help coordinate the glycine carboxyl group. We therefore introduced the corresponding substitutions into 15599 Pplase2. As predicted from the single-mutant defects, the G227K/D238Y double mutant lost activity with 1,3-diaminopropane (Fig. 3C and F). However, when glycine was supplied as donor, the same variant generated glyT from 5hmdU-DNA (Fig. 3F and G). Thus, replacement of only two active-site residues was sufficient to redirect donor preference from diamine transfer to glycine transfer, demonstrating that ligand selectivity in Pplase2 is specified by a minimal set of pocket determinants and can be rationally reprogrammed. The diversity at ligand-binding positions suggests that the Pplase2 family likely recognizes many additional donor substrates that remain to be identified (Supplementary Fig. S6B).

Structural modeling explains the substrate range and 1,3-diaminopropane preference of 15599 Pplase2

To explore the basis for the broad substrate scope but strong preference of 15599 Pplase2, we modeled the enzyme in complex with Mg²⁺, 5-PmdU-DNA, and a series of linear diamines with chain lengths ranging from C2 to C7, using the YuA structure as a template. The model indicates that the binding pocket can accommodate linear diamines from C2 to C6, consistent with the experimentally observed substrate range (Fig. 2A and B). However, the bound conformations of diamines with chain length longer than that of 1,3-diaminopropane all bear high ligand conformational strain energies (Supplementary Fig. S7A–E): C4 to C6 diamines have strain energies as high as ∼8–9 kcal/mol, while C2 and C3 diamines have negligible strain energies. Together, these results provide a plausible explanation for the defined substrate range of 15599 Pplase2 and its strong preference for 1,3-diaminopropane.

Comparison with PhiW-14 Pplase2 suggested that M18, which replaces an Ile at the corresponding position, may further restrict pocket volume and favor shorter ligands (Supplementary Fig. S7F). However, neither the corresponding single mutant nor the M18X/A88E double mutant could be purified, precluding direct experimental testing.

Taken together, these data define the functional architecture of the Pplase2 active site (Fig. 3H). The invariant cysteine acts as the catalytic residue, whereas D25, E222, and T223 are required for divalent metal coordination. Residues including M18, G227, H230, D238, and D243 shape the ligand-binding pocket and influence donor selectivity. Among these, G227, H230, and D238 are rather conserved within individual Pplase2 clusters but differ between clusters, identifying them as likely determinants of small-molecule specificity (Supplementary Fig. S6B). The successful conversion of 15599 Pplase2 into a glycine-utilizing enzyme by pocket remodeling further supports this view and establishes Pplase2 as an engineerable family of DNA modification enzymes.

A catalytic model for dapT formation by 15599 Pplase2

Based on the catalytic mechanism proposed for Pplase2-catalyzed glyT formation [12] and our mutagenesis analyses of 15599 Pplase2 (Fig. 3), we propose a catalytic model for dapT biosynthesis (Supplementary Fig. S7G). The reaction is initiated by activation of 5-PmdU by the strictly conserved catalytic residue C226, followed by phosphate elimination at the C5 hydroxymethyl group to generate an exocyclic methylene-like intermediate. The amino group of 1,3-diaminopropane then attacks this intermediate to yield dapT. Our mutational analyses further indicate that E222 and T223 coordinate the divalent metal ion required for catalysis. This metal ion likely contributes both to substrate binding and to stabilization of the reaction intermediate, thereby promoting catalysis.

DapT naturally occurs in the genome of phage 15599

Having established that 15599 Pplase2 generates dapT in vitro, we next asked whether this modification is present in the phage genome. Genomic DNA from phage 15599 was isolated, hydrolyzed to nucleosides, and analyzed by LC-MS/MS. Alongside the four canonical nucleosides, dA, dT, dC, and dG, we detected dapT in the DNA hydrolysate (Fig. 4A and B), establishing dapT as a natural DNA modification in phage 15599.

Figure 4.

DapT is a natural DNA modification in phage 15599, with host-derived 1,3-diaminopropane serving as its biosynthetic donor. (A) HPLC traces of nucleosides from digested phage 15599 genomic DNA. Standard nucleosides are shown for comparison. (B) MS/MS spectra of dapT from digested 15599 genomic DNA and synthetic dapT. (C) Genomic neighborhoods of Pplase2 in Acinetobacter phage SH-Ab 15599 and Delftia phage PhiW-14. (D) Proposed pathway for thymine hypermodification in phage 15599.

Quantification of canonical nucleosides by UV absorbance yielded average dC/dG and dT/dA ratios of 0.98 and 0.77, respectively, indicating ∼23% thymine residues harbor dapT modification. Given the 62% AT content of the phage 15599 genome, we estimate that dapT constitutes ~14% of total nucleosides.

Although 15599 Pplase2 accepts multiple diamine substrates in vitro, only dapT was detected in phage DNA (Figs 2A and B; and 4A and B), suggesting a strong in vivo preference for 1,3-diaminopropane. This interpretation is consistent with previous cation-exchange HPLC measurements showing that intracellular 1,3-diaminopropane is the predominant polyamine in A. baumannii, whereas putrescine is nearly absent in this species [39, 40] Together, these findings indicate that the exclusive occurrence of dapT in phage 15599 genome is driven by both the intrinsic substrate preference of 15599 Pplase2 and the high cellular abundance of 1,3-diaminopropane in the host.

A biosynthetic pathway for dapT in phage SH-Ab 15599 and its host

To identify enzymes involved in dapT biosynthesis, we analyzed the genomes of phage 15599 and its host. In addition to the previously characterized 5-HMUDK and Pplase2, the phage genome encodes a DctD, a dUMP HMase, and a DNA polymerase (Fig. 4C and Supplementary Fig. S8). Together with established thymine hypermodification pathways [2, 12], these findings allowed us to reconstruct a pre-replicative route to 5hmdU-DNA and a post-replicative route to dapT (Fig. 4D).

In the proposed pre-replicative pathway, dCMP is first deaminated to dUMP by DctD, and dUMP is then converted to 5hmdUMP by the phage-encoded dUMP HMase. Subsequent phosphorylation by host dNMP/dNDP kinases would generate 5hmdUTP, which is then incorporated into phage DNA by the phage DNA polymerase to produce 5hmdU-DNA. In the post-replicative step, 5-HMUDK phosphorylates the C5 hydroxymethyl group of 5hmdU in DNA to generate a reactive phosphoester intermediate. This intermediate is then converted by 15599 Pplase2, using host-derived 1,3-diaminopropane as the donor, to yield dapT (Fig. 4D). Together, these results define a complete biosynthetic model for dapT formation in phage 15599.

A bifunctional dTTPase/dUTPase channels nucleotide metabolism toward dapT biosynthesis

Previous work on phage PhiW-14 showed that infection induces dTTPase activity and consequently depletes cellular dTTP [43]. We therefore searched the 15599 genome for enzymes that might similarly remodel nucleotide pools and identified an uncharacterized protein, gp18 (NCBI accession: AXF41387.1), containing a C-terminal domain with a dUTPase/dCTP pyrophosphatase-like fold.

To test its function, we purified gp18 and assayed its activity in vitro (Supplementary Fig. S2). Phosphomolybdate colorimetry and LC-MS/MS showed that gp18 selectively hydrolyses dTTP and dUTP to dTMP and dUMP, respectively, with no detectable activity toward 5hmdUTP, other dNTPs, or NTPs (Supplementary Fig. S9A–E). Hydrolysis was detected in the colorimetric assay only in the presence of inorganic pyrophosphatase, indicating release of pyrophosphate (Supplementary Fig. S9F). Metal-ion screening further showed that Co²⁺ supports maximal activity (Supplementary Fig. S9G). Given that homologous dUTPase exist as dimers in solution [44], we modeled the C-terminal dTTPase/dUTPase domain as a dimer in complex with dUTP, dTTP, dCTP, or 5hmdUTP to understand the basis for substrate selectivity. In the dUTP- and dTTP-bound models, Q13, N17, and Y35 form hydrogen bonds with the pyrimidine base. Conversely, both dCTP and 5hmdUTP are predicted to experience steric clashes with W56 from the opposing protomer, thereby disfavoring their accommodation within the active site (Supplementary Fig. S9H–K). We therefore designate gp18 as a bifunctional dTTPase/dUTPase. This activity is well suited to the proposed dapT pathway. Hydrolysis of dTTP would reduce competition with 5hmdUTP during phage DNA replication, favoring 5hmdU incorporation, whereas hydrolysis of dUTP to dUMP would replenish substrate for 5hmdUTP biosynthesis (Fig. 4D). Together, these data identify gp18 as a pathway-linked dTTPase/dUTPase that redirects nucleotide metabolism toward dapT biosynthesis.

Incorporation of phage-derived dapT into the E. coli genome

In addition, we co-expressed the key enzymes required for dapT formation in E. coli, including the DctD, dUMP HMase, dTTPase/dUTPase, 5-HMUDK, and Pplase2. Previous studies have shown that, in mid-log-phase E. coli B or K-12 grown in minimal medium at neutral pH, putrescine (∼13.1 μmol/g wet weight; ∼0.02 M) and spermidine (∼4.7 μmol/g; ∼0.006 M) are the predominant intracellular polyamines, whereas no other diamines or polyamines (including spermine) are detectable in either the cells or the culture medium under these conditions [45]. Consistent with this, the modified nucleoside dapT was detected in the E. coli genome only when all of the key enzymes were co-expressed and 1,3-diaminopropane was additionally supplied in the culture medium. In contrast, when no exogenous 1,3-diaminopropane was added, overexpression of the pathway enzymes led to the utilization of the endogenous E. coli metabolite putrescine, resulting in the formation of putT, the same thymine hypermodification found in PhiW-14 (Supplementary Fig. S10). These results demonstrate that these proteins are essential for the biosynthesis of this new thymine hypermodification and further highlight their potential as versatile enzymatic tools for installing unusual DNA modifications.

Sequence specificity of thymine hypermodification in the phage 15599 genome

Despite major advances in high-throughput sequencing, current second- and third-generation sequencing platforms remain limited in their ability to reliably identify non-canonical nucleoside modifications in genomes. We therefore used a panel of REases recognizing T-containing sequences to map thymine hypermodification in the phage 15599 genome.

We first tested 10 REases with 4-bp recognition sites (Fig. 5A). Phage 15599 genomic DNA was completely resistant to NlaIII (CATG), highly resistant to HpyCH4V (TGCA; only a trace amount of DNA was cleaved), and also showed strong resistance to MboI (GATC). In contrast, only weak resistance was observed for TaqI (TCGA), BfaI (CTAG), RsaI (GTAC), HpyCH4IV (ACGT), and AluI (AGCT), whereas the genomic DNA was completely digested by enzymes recognizing AATT or TTAA (Fig. 5A). These results suggest that dapT occurs preferentially in TG dinucleotide contexts, to a lesser extent in TC contexts, and much less frequently in TT or TA contexts.

Figure 5.

The TG dinucleotides are highly modified in the genome of phage 15599. (A) Restriction digestion of phage genomic DNA with enzymes recognizing 4-bp sequences. (B) Restriction digestion of phage genomic DNA with enzymes recognizing TG-containing sequences. Left, predicted banding patterns for an unmodified phage genome digested with the indicated REases; middle, observed gel electrophoresis patterns after digestion with the same enzymes; right, recognition sequences of the restriction endonucleases and the numbers of their sites in the phage genome. R, resistant; VP, very partial digestion; SP, slight partial digestion; LP, extensive partial digestion; and C, complete digestion.

To further test this hypothesis, we treated phage 15599 genomic DNA with 15 additional restriction endonucleases recognizing TG-containing sequences (Fig. 5B). The DNA was completely resistant to 14 of the 15 enzymes and remained highly resistant to MfeI (CAATTG), indicating that TG dinucleotides within these recognition sites are almost fully occupied by dapT hypermodification. Analysis of the phage 15599 genome showed that TG dinucleotides account for ∼13.8% of all dinucleotide sites, closely matching the estimated genomic abundance of dapT (∼14%). Together, these data support the conclusion that thymine hypermodification in phage 15599 is sequence-specific and occurs predominantly at TG sites.

Consistent with these findings, previous restriction endonuclease analyses of thymine-hypermodified phage genomes showed that hypermodification occurs mainly at TG or TC sites [8]. Notably, the PhiW-14 genome was reported to be completely resistant to NlaIII and HpyCH4V digestion [8], a pattern highly similar to that observed for phage 15599 (Fig. 5A), suggesting that the two phages share the same sequence preference for thymine hypermodification.

Role of thymine hypermodification in evasion of the host restriction-modification system

Many bacteriophages incorporate modified nucleosides at high frequencies, rendering their genomes partially or completely resistant to cleavage by restriction endonucleases. Previous studies of phages M6, ViI, and PhiW-14, all of which carry thymine hypermodifications, showed that their genomic DNAs are resistant to 48.4%, 71.0%, and 68.8%, respectively, of >200 commercially available type II restriction enzymes [8].

To test whether thymine hypermodification protects the 15599 genome from host restriction endonucleases, we searched UniProt and identified homologs of six restriction enzymes in its host (Supplementary Table S3). With the exception of HpaII, the recognition sequences of the other five enzymes all contain T and fall into three groups. BsuBI, PstI, and AplI recognize the same sequence, CTGCAG. As this sequence contains the HpyCH4V motif TGCA, and only 21 CTGCAG sites are present in the 15599 genome, we reasoned that 15599 DNA would resist PstI digestion, which was confirmed experimentally (Fig. 5B). The other two enzymes, EcoRII and CfrBI, recognize CCWGG and CCWWGG, respectively. Digestion with their isoschizomers BstNI and StyI showed that the 15599 genome is completely resistant to cleavage at both sites (Fig. 5B). Together, these results indicate that the thymine hypermodification dapT protects phage 15599 against the restriction-modification system of A. baumannii.

Phage infection redirects host 1,3-diaminopropane biosynthesis to supply the modification precursor

To define the transcriptional landscape of phage 15599 and the host response to infection, we performed time-resolved RNA-seq on A. baumannii collected at multiple time points after infection. Mapping reads to both the phage and host genomes revealed a highly coordinated temporal pattern of gene expression (Supplementary Fig. S11A and B). Phage genes were classified into four functional categories: structural, non-structural, lysis, and hypothetical. Temporal analysis showed that non-structural genes were predominantly expressed early during infection, whereas structural genes were strongly induced at later stages (Fig. 6A and Supplementary Fig. S11C), consistent with the canonical transcriptional program of lytic phages.

Figure 6.

Transcriptional dynamics of phage 15599 and infection-induced metabolic reprogramming in its host A. baumannii. (A) Hierarchical clustering heatmap showing temporal expression profiles of phage 15599 genes post-infection. Expression levels are shown as row-scaled Z-scores of log₂-transformed TPM values, ranging from blue (low) to red (high). The sidebar indicates gene categories. Genes involved in dapT biosynthesis are labeled. (B) Heatmap showing differential expression of A. baumannii genes involved in l-aspartate and 1,3-diaminopropane metabolism. Colors indicate log₂ fold changes (Log₂FC) relative to the uninfected control (0 min), with red and blue denoting upregulation and downregulation, respectively. Asterisks indicate statistical significance, q-value determined by DESeq2: *, q < 0.05; **, q < 0.01; ***, q < 0.001. (C) Schematic of the 1,3-diaminopropane biosynthetic pathway and competing branches in A. baumannii. l-Asp is converted to 1,3-diaminoporpane via ASA and diaminobutyrate (DAB), catalyzed by Ask, Asd, Dat, and Ddc. Competing pathways consuming ASA, catalyzed by Hom and DapA, and acetylation of 1,3-diaminopropane by Dpa are also shown. (D) Quantification of biofilm biomass after 24 h of phage treatment at different MOIs. (E) Enumeration of viable bacteria in the culture medium by CFU assay after 24 h of phage treatment at different MOIs. ^****P < 0.0001 versus the control group.

Concurrently, infection significantly reprogrammed host metabolism, particularly the pathway responsible for synthesis of 1,3-diaminopropane, the precursor of the phage-specific DNA modification dapT (Fig. 6B). In A. baumannii, dap is synthesized from l-aspartate through a pathway involving l-aspartyl-4-semialdehyde dehydrogenase (Asd), diaminobutyrate aminotransferase (Dat) and 2,4-diaminobutyrate decarboxylase (Ddc) (Fig. 6C). Under normal conditions, dap contributes to surface motility, but can be depleted by acetylation through diaminopropane acetyltransferase (Dpa), a process associated with enhanced bacterial adherence [46]. In addition, l-aspartate 4-semialdehyde (ASA) is also consumed by homoserine dehydrogenase (Hom) and 4-hydroxy-tetrahydrodipicolinate synthase (DapA) in essential amino acid biosynthesis [47].

Differential expression analysis of genes involved in 1,3-diaminopropane metabolism, using the onset of infection as the reference, showed that the three biosynthetic genes asd, dat, and ddc were upregulated following infection. In contrast, hom and dapA, which divert ASA into competing pathways, as well as dpa, which depletes 1,3-diaminopropane, were downregulated to varying extents (Fig. 6B). These data indicate that phage 15599 infection redirects host metabolic flux toward 1,3-diaminopropane production, thereby increasing availability of the precursor required for phage DNA hypermodification.

To provide a broader comparison, we examined the temporal expression dynamics of essential host genes post-infection, selected from a previously defined essential gene set [48] (Supplementary Fig. S12 and Supplementary Table S4). We observed that phage infection triggered a coordinated upregulation of genes associated with the respiratory chain and ATP synthase, and genes of tricarboxylic acid cycle were concurrently upregulated to promote NADH production, effectively fueling ATP synthesis. The sustained expression of ribosomal proteins implies that this bioenergetic metabolic shift is driven by the translational demands of viral protein synthesis. Meanwhile, pathways involved in capsular polysaccharide production, host DNA replication, nucleotide metabolism, and riboflavin biosynthesis underwent downregulation to conserve metabolic resources for phage propagation, consistent with previous studies [49].

We further examined the effect of phage infection on host biofilm formation at different MOIs. After 24 h, phage 15599 significantly inhibited biofilm formation under all conditions tested, while viable cell counts showed a marked reduction in bacterial growth following infection (Fig. 6D and E; and Supplementary Fig. S13). Thus, the decrease in biofilm formation is likely to be largely attributable to impaired bacterial growth, although additional effects on biofilm-associated pathways cannot be excluded.

Discussion

Our study expands the chemical repertoire of Pplase2-dependent DNA hypermodifications by identifying dapT, a novel thymine modification formed through the conjugation of host-derived 1,3-diaminopropane to 5hmdU in phage 15599 DNA. Together with putT in phage PhiW-14, dapT represents a rare example of a natural DNA modification derived from a host polyamine substrate [50]. Moreover, the dapT pathway exists beyond phage 15599. Acinetobacter phage nACB2, which clusters with phage 15599 in the SSN, encodes a related Pplase2 homolog that retains the key catalytic residues and shares an overall sequence identity of 49% with 15599 Pplase2, exhibits a similar genomic organization to that of phage 15599 (Supplementary Figs S6A and S8), and infects the same bacterial genus [14, 51], suggesting that it also harbors the dapT modification.

Our discovery points to an adaptation of Pplase2-dependent DNA hypermodification to host metabolism. Polyamine composition varies across bacterial species, and the dominant polyamines in D. acidovorans and A. baumannii correspond to the substrates used by their phages PhiW-14 and 15599, respectively [40, 52]. This suggests that Pplase2 enzymes evolve within the constraints of host metabolite availability while maintaining substrate selectivity. Consistent with this, 15599 Pplase2 accepts diamines of varying chain lengths in vitro, yet still generates the naturally occurring dapT modification in the phage genome (Figs 2 and 4; and Supplementary Fig. S4). Notably, its substrate preference can be reprogrammed by active-site mutations (Fig. 3F and G), revealing the catalytic and structural plasticity of the Pplase2 family. The reactive amino group retained in this DNA modification can be exploited for downstream labeling via click chemistry or related reactions, offering potential applications in DNA labeling, enrichment, imaging, and modification mapping. Together, these results identify Pplase2 as a versatile and engineerable family of DNA hypermodifying enzymes that hold the promise for generating desired DNA modifications.

Polyamines are positively charged molecules that stabilize nucleic acids through electrostatic interactions and also regulate biofilm formation [53, 54]. Because adherence is the first step in biofilm development, changes in polyamine abundance or modification state may influence the transition between motile and surface-associated lifestyles. In A. baumannii, the balance between free and Dpa-acetylated 1,3-diaminopropane appears to coordinate motility and biofilm formation: non-modified 1,3-diaminopropane is associated with enhanced motility, whereas Dpa activity promotes surface adherence and biofilm development [46]. In this study, RNA-seq analysis showed that phage 15599 infection upregulated host 1,3-diaminopropane biosynthesis, presumably to supply precursors for phage thymine hypermodification. Phage 15599 also reduced host biofilm biomass, but this effect coincided with strong growth inhibition, suggesting that the biofilm defect is at least partly attributable to reduced bacterial abundance. Nevertheless, phage-mediated rewiring of 1,3-diaminopropane metabolism may increase the pool of free 1,3-diaminopropane while limiting its acetylated form, thereby disfavoring early surface adherence. Whether this metabolic shift directly affects biofilm formation independently of growth inhibition remains to be determined.

The host of phage 15599, A. baumannii, is a member of the ESKAPE pathogens —Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, A. baumannii, Pseudomonas aeruginosa, and Enterobacter species—a group of clinically important bacteria associated with multidrug-resistant infections. In particular, carbapenem-resistant A. baumannii represents an urgent clinical threat [55]. Amid the escalating crisis of antibiotic resistance, phage therapy has re-emerged as a promising antibacterial strategy. Although several successful applications and clinical trials have been reported in recent years, most remain at an early stage [56–58]. In this context, our characterization of thymine hypermodification in phage 15599 provides a conceptual framework for phage engineering and may support future development of phage therapy.

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

None declared.

Funding

This work was supported by grants from the National Natural Science Foundation of China [32370704 to S.Z.], the Science and Technology Commission of Shanghai Municipality [25ZR1402366 to W.L.], the ShanghaiTech University School of Life Science Development Fund [S.Z.], and the AI4S Program at ShanghaiTech University [S.Z.]. Funding to pay the Open Access publication charges for this article was provided by the ShanghaiTech University School of Life Science Development Fund.

Data availability

The crystal structure of YuA Pplase2 in complex with 5hmdU-DNA and glycine has been deposited with the Protein Data Bank under DOI https://doi.org/10.2210/pdb25qc/pdb. The whole genome shotgun sequence of A. baumannii strain 15599 has been deposited in the NCBI database under BioProject accession number PRJNA1430119 and in DDBJ/ENA/GenBank under accession number JBXTSJ000000000. The version described in this paper is JBXTSJ010000000. The RNA-seq data were deposited in GEO under accession GSE322957. The deposited data are publicly available as of the date of publication. Other data are available from the corresponding author upon reasonable request.