Unraveling the Prevalence and Multifaceted Roles of Accessory Peptide Deformylases in Bacterial Adaptation and Resistance

Abstract

Peptide deformylases (PDFs) are enzymes that are essential for bacterial viability and attractive targets for antibiotic development. Yet, despite their conserved function, many bacteria encode multiple PDFs, a genomic feature whose prevalence and implications remain largely unexplored. Here, we reveal that nearly half of all bacterial genomes carry more than one PDF gene, frequently embedded within mobile genetic elements such as plasmids and integrons. In Vibrio cholerae, the accessory PDF (Def2_VCH) confers reduced susceptibility to actinonin (ACT), the most studied PDF inhibitor, while still supporting bacterial growth in the absence of the canonical PDF copies (Def1_VCH). Crystallographic analysis shows that this reduced susceptibility stems from an arginine-to-tyrosine substitution that probably reduces ACT binding. Strikingly, this resistance signature is shared by integron-encoded PDFs, and transfer of an integron-encoded PDF cassette from Pseudoxanthomonas into a susceptible V. cholerae is sufficient to abolish ACT susceptibility. These findings reveal a hidden reservoir of resistance within the bacterial mobilome and shed light on potential mechanisms of bacterial resilience to environmental PDF inhibitors.

Introduction

In bacteria, as well as in mitochondria and chloroplasts, translation is always initiated by a formylated methionine, which is usually cleaved during translation (reviewed in Giglione et al. 2015). Peptide deformylases (PDFs) are metalloproteases that catalyze the removal of the formyl group from the initiator methionine, a critical step in the N-terminal methionine excision (NME) process (Mazel et al. 1994; Giglione et al. 2015). PDFs are highly promiscuous enzymes with limited specificity, deformylating approximately 95% of the bacterial proteome (Bienvenut et al. 2015). Generally, only membrane or exported proteins, which have a hydrophobic signal peptide at their N-terminus, escape deformylation (Yang et al. 2022).

PDFs are relatively small enzymes, (150 to 200 amino acid (aa) in prokaryotes and 220 to 250 aa in eukaryotes) with low sequence identity (20% to 30%) but significant structural conservation (Giglione et al. 2004). They mainly differ in the structure of their C-terminal region, which has been shown in Escherichia coli to facilitate interaction with the ribosome (Bingel-Erlenmeyer et al. 2008; Sandikci et al. 2013; Bhakta et al. 2019; Akbar et al. 2021). Most PDFs share a common catalytic core defined by three conserved sequence motifs (I, II, and III), which coordinate a metal ion and position key residues for substrate processing (Giglione et al. 2004).

Phylogenetically, PDFs are classified into four main groups called types (Grzela et al. 2017). Type 1A and 1B PDFs, found in both eukaryotes and prokaryotes, have a characteristic C-terminal tail folded into an α-helix (Chan et al. 1997; Giglione et al. 2009). Type 2 PDFs are mostly found in Gram-positive bacteria and have a C-terminal tail folded into a β-strand (Giglione et al. 2009; Grzela et al. 2017). The structures of type 3 PDFs, as well as of type 4 PDFs, have not yet been resolved. Type 3 PDFs, present in Archaea, protists, and some bacterial genera, display substitutions in conserved motifs I and II, which have been shown to lead to reduced enzymatic activity or even inactive enzymes (Margolis et al. 2000; Giglione et al. 2004; Bouzaidi-Tiali et al. 2007; Grzela et al. 2017). The type 4 PDFs remain less well-characterized and are found in a broad variety of hosts, including marine viral bacteriophage and eukaryotic parasites (Grzela et al. 2017).

Given their conservation and indispensable roles in the bacterial translation process, PDFs have been proposed as attractive targets for the development of new antibiotics (Mazel et al. 1994). Actinonin (ACT), a natural product made by bacteria from the genus Streptomyces (Gordon et al. 1962, 1975), was identified as the first PDF inhibitor (Chen et al. 2000). Since then, numerous other PDF inhibitors have been discovered (reviewed extensively in Jain et al. 2005; Sangshetti et al. 2015). All peptide-based PDF inhibitors share a common mechanism of action: they bind to the catalytic pocket of the enzyme, thereby preventing substrate access and inhibiting its deformylase activity (Sangshetti et al. 2015). To date, three mechanisms of reduced susceptibility to ACT have been described: (1) drug export by efflux pumps (Chen et al. 2000; Margolis et al. 2000; Clements et al. 2001); (2) impairment of the formylation mechanism by inactivation of fmt, the gene which encodes the methionyl-tRNA formyltransferase (FMT), that directly catalyzes the formylation of initiator methionine, or of folD or glyA, whose products contribute to the synthesis of the formyl group donor essential for FMT activity (Margolis et al. 2000; Nilsson et al. 2006; Duroc et al. 2009; Yang and Sun 2016); and (3) overexpression or mutations of the PDF (Margolis et al. 2001; Dean et al. 2007). Although inactivation of the formylation process is the most common mechanism of reduced susceptibility, it significantly slows bacterial growth (Guillon et al. 1992; Mazel et al. 1994) and could lead to a drastic reduction in virulence, as shown for Staphylococcus aureus in mouse models (Margolis et al. 2000), indicating that despite the development of resistance, PDF inhibitors may remain effective in treating bacterial infections due to the co-associated growth defect and reduced virulence.

For over two decades, it has been observed that some bacteria possess multiple PDF encoding genes while they contain only one fmt gene, allowing the formylation of the initiator methionine (Margolis et al. 2000; Guilloteau et al. 2002). Notable species among these include important model organisms or pathogens such as S. aureus, Enterococcus faecalis, Bacillus subtilis, Streptococcus pneumoniae, Vibrio cholerae, and Pseudomonas aeruginosa (Margolis et al. 2000). In certain cases, such as S. aureus and S. pneumoniae, substitutions in the conserved motifs essential for catalytic activity result in the inactivation of one of the PDFs in the laboratory conditions tested (Margolis et al. 2000, 2001). In other examples, such as in B. subtilis, both PDFs are active and share similar enzymatic and inhibition properties (Haas et al. 2001). Despite the longstanding observation of multiple PDF genes, no large-scale genomic analysis has been performed to study the distribution of PDFs in bacterial genomes, and the evolutionary advantage of expressing several different PDFs remains unclear. One hypothesis is that this redundancy could be linked to resistance against PDF inhibitors, natural products synthesized by bacteria that may be encountered in diverse environmental contexts. Moreover, with the rise of antibiotic resistance worldwide, it becomes urgent to identify new therapeutic targets, and as PDFs have been shown to be a promising target, understanding the distribution and role of these accessory PDFs is crucial.

To address this knowledge gap, we explored the distribution of PDF across RefSeq bacterial genomes and focused on the functional roles of some specific PDFs, including those found in Vibrio species. Our findings reveal that approximately half of the bacterial reference genomes possess multiple PDF genes, with significant intra-species variability. We discovered that PDFs can be encoded on mobile genetic elements (MGEs), such as plasmids and integrons, which likely contribute to this variability and suggest an adaptive role. Furthermore, our investigation of V. cholerae, which encodes two type 1b PDFs with 50% sequence homology (Def1_VCH and Def2_VCH), demonstrated that both PDFs are catalytically active. We found that Def2_VCH specifically confers reduced susceptibility to the natural PDF inhibitor, ACT. Additionally, the integron PDF cassette identified in a Pseudoxanthomonas suwonensis strain was shown to impart significant resistance to PDF inhibitors in both V. cholerae and E. coli. This resistance phenotype suggests that the dissemination of this cassette could pose a challenge for the therapeutic use of PDF inhibitors and raises questions about their viability in clinical applications.

These discoveries underscore the diverse distribution of PDF genes in bacteria and support the hypothesis that many additional PDF gene copies have been selected for their role in resistance to natural PDF-targeting antibiotics, thereby questioning the potential for clinical application of PDF inhibitors in treating bacterial infections.

Results

Half of Bacterial Species Harbors Several PDF

Despite longstanding knowledge of PDF duplication, no large-scale study has explored the distribution of PDFs across bacterial genomes. We analyzed 5,042 fully sequenced genomes of bacterial reference species available in the RefSeq database as of May 4, 2024, and measured the occurrence of genes coding for PDFs in their genomes (Fig. 1a). PDF genes were extracted using annotations from the NCBI Prokaryotic Genome Annotation Process (PGAP). Annotation accuracy was confirmed by BLASTp and HMMER searches (profile PF01327), leading to similar results.

Fig. 1.

Prevalence and distribution of accessory peptide deformylases in bacterial genomes. a) Distribution of Peptide deformylases in bacterial reference genomes. The number of species is indicated above the histograms. b) Amino acid conservation is shown for the three motifs originally described by (Giglione et al. 2004) and for revised motifs derived from PDFs found in bacterial reference genomes encoding a single PDF copy. The overall conservation of each motif across all PDFs is indicated below, along with the proportion of PDFs that exactly match the canonical or updated motif sequences. Conservation values for individual amino acid positions are indicated above or below the corresponding residue, based on the subset of genomes encoding a single PDF. Amino acid conservation below 95% is highlighted in red. Φ represents a hydrophobic amino acid and X represents any amino acid. c) Violin plot illustrates the size distribution of plasmids encoding PDFs. The violin plot shows the density of plasmid sizes, with the white box inside representing the interquartile range (IQR) and the central line indicating the median size. d) Bar chart depicting the classification of PDF-encoding plasmids based on their mobility: 36% encode all the key genes required for conjugation, i.e., mating pair formation (MFP) machinery and a relaxase (conjugative plasmids: pCONJ), 2% encode an incomplete MFP machinery and a relaxase (decayed conjugative plasmids: pdCONJ), 3% encode only a relaxase (mobilizable plasmids: pMOB), and 31% encode an oriT without relaxase (pOriT). e) Phylogeny and genomic context of PDFs encoded within integron cassettes. The phylogenetic tree (left) represents the evolutionary relationships among PDF protein sequences encoded within integron cassettes. The bar plot (right) represents the total number of cassettes in each integron, with colors distinguishing complete integrons (blue) from CALINs (yellow). The relative position of the PDF cassette within each integron is indicated by a cross when it corresponds to the sequence used in the phylogeny and by a circle when additional PDF cassettes are present in the same integron. The phylogenetic tree was constructed using IQ-TREE multi-core v.2.2.2.2 (Minh et al. 2020) based on a MAFFT-aligned dataset (Katoh and Standley 2013). Among 1,232 proteins models tested, the Q.plant + G4 model was selected as the best-fitting model according to the Bayesian information criterion using the IQ-TREE-M TESTNEW algorithm. Statistical support was assessed using 1,000 ultrafast bootstrap replicates and 1,000 SH approximate likelihood ratio test replicates (Hoang et al. 2018).

This analysis revealed that 48% of bacterial species harbor more than one PDF. Notably, over 2.7% of reference genomes contain more than three PDF genes, with Dactylosporangium vinaceum holding the record with seven PDFs. Among the 5,042 reference genomes, only 31 genomes lacked identifiable PDF genes. Of these, 29 are endosymbiotic bacteria, where the absence of essential genes has been previously documented (McCutcheon and Moran 2011), and 2 are probably incomplete as they also lack the formyltransferase fmt gene thought to be essential and conserved among all bacterial species (Table S1).

Conservation of Essential Residues: Refinement of the Catalytic Motifs

The PDF sequence contains three conserved motifs reported to be essential for enzymatic activity: motif I (GΦGΦΑAXQ), motif II (EGCXS), and motif III (QHEXDHLXG), where Φ represents a hydrophobic amino acid, and X represents any amino acid (Giglione et al. 2004).

To evaluate the potential activity of these PDFs, we analyzed the presence of these three conserved motifs (Giglione et al. 2004). Surprisingly, only 61.9% of the PDFs exhibited exact conservation of these motifs. This may suggest that many accessory PDFs might be weakly active or inactive, similar to those identified in S. aureus and S. pneumoniae and belonging to the type 3 PDF (Margolis et al. 2000, 2001; Grzela et al. 2017). However, 24.5% of the reference genomes did not encode any PDF matching the exact sequence of these motifs. Given that at least one active PDF is essential for bacterial survival, this suggests that the motifs described by Giglione and colleagues may not be reliable enough as indicators of PDF activity across all bacterial species.

Aligning PDFs from genomes harboring only one PDF revealed that the motifs described by C. Giglione et al. (Giglione et al. 2004) are indeed not strictly conserved (Fig. 1b).

Based on these observations, we propose an update to the motifs described by C. Giglione and colleagues, with a cut-off for the amino acids conserved in more than 95% of PDFs present in genomes with only a single PDF copy: motif I (ΦGΦΑAXQ), motif II (EGCLSΦ), and motif III (HEXXHΦXG). The majority of PDFs in bacterial genomes (80%) possess these updated motifs, encompassing the amino acids critical for enzymatic activity (Fig. 1b). However, 7.1% of bacterial reference genomes still do not encode a PDF with the exact sequence of these updated motifs indicating that the exact conservation of these three motifs is not essential for the PDF enzymatic activity, or that they contain sequencing errors.

Conservation and Genomic Organization of Formylation and Deformylation Genes in Bacteria

Although it is widely assumed that the formylation of initiator tRNA^Met by Methionyl-tRNA formyltransferase is a conserved feature of bacterial translation, to our knowledge, no recent large-scale study has systematically demonstrated this conservation. Similar to our analysis of PDFs, we screened 5,042 bacterial reference genomes and found that the vast majority (98.6%) encode a single copy of the fmt gene. Only ten genomes encode two fmt homologs, while the few genomes lacking fmt (73 in total) predominantly correspond to obligate endosymbionts with highly reduced genomes.

It is also generally accepted that fmt and def genes are co-localized and often organized into a single operon, as exemplified in E. coli (Meinnel and Blanquet 1993; Mazel et al. 1994, 1997). Genomic context analysis of PDF genes revealed that, although fmt and def are frequently co-localized, this association is conserved in only 58% of bacterial species. Furthermore, in E. coli, the fmt-PDF operon additionally includes rsmB and trkA, encoding proteins involved in 16S rRNA methylation and potassium transport regulation, respectively. This extended operonic structure is restricted to Gammaproteobacteria and found in only 62% of reference species within this clade.

Additionally, we identified 145 bacterial species in which PDF genes are located in close proximity to other PDF genes (less than 5 genes apart), consistently on the same DNA strand, suggesting potential co-transcription. Notably, 104 of the 145 species exhibiting clustered PDF genes were found to belong to the order Rhodobacterales. This genomic organization appears to be a conserved feature of this clade, as 98.1% of the Rhodobacterales reference species in our dataset harbor multiple PDF genes arranged in a cluster. In the majority of these Rhodobacterales genomes (62.5%), three PDF genes were found grouped together. Although highly conserved in Rhodobacterales, the biological rationale behind the clustering of PDF genes remains elusive.

An unusual case was observed in the four reference genomes of Exiguobacterium, a genus from the order Bacillales, in which def and fmt appear to be fused into a single gene. Interestingly, all Exiguobacterium genomes also encode a second PDF containing all residues critical for catalytic activity. To assess the functionality of the def-fmt fusion, we synthesized a codon-optimized version of the fused CDS and expressed it in an E. coli Δdef-fmt mutant (Mazel et al. 1994). Complementation restored normal growth to levels comparable to the WT strain, demonstrating that the fusion protein retains both deformylase and formyltransferase activities (Figure S1). These results confirm that the def-fmt fusion is functional, suggesting that it likely represents a genuine biological feature rather than a sequencing or annotation artifact.

PDFs are Encoded on Mobile Genetic Elements

The analysis of all complete bacterial genomes in RefSeq as of April 28, 2024 (40,298 genomes), distributed among 5,728 species, revealed intra-species variability in the number of PDFs encoded within bacterial genomes. Indeed, approximately 10% of bacterial species with at least two complete genomes in this database show no conservation in PDF copy number, suggesting that some PDFs may be associated with mobile genetic elements (MGEs). To further investigate this possibility, we analyzed a plasmid database (PLSDB) and an integron cassette database to determine whether PDFs are encoded within these MGEs.

Plasmid

In the plasmid database, we identified 676 replicons harboring at least one PDF gene. However, this database also misincludes some secondary chromosomes or chromids. Generally, chromosomes and chromids are distinguished from plasmids by their “essential” nature and larger size (Harrison et al. 2010; diCenzo and Finan 2017). Applying an additional filter to remove replicons larger than 1 Mb, we identified 657 replicons smaller than 1 Mb, carrying one PDF gene and 10 harboring 2 PDF genes (Table S2). Analysis revealed that the median size of plasmids encoding a PDF gene is 210 kb (Fig. 1c). They are mainly found in K. pneumoniae (57.2%) and E. coli (25.9%), but they can be found among 58 different bacterial species from 35 genera (Figure S2a and Table S2). A significant proportion (67.5%) of the identified plasmidic PDFs are identical to the chromosomal E. coli PDF, typically found in an operon with fmt, and are distributed across 12 Enterobacteriaceae species spanning the genera Klebsiella, Escherichia, Shigella, Salmonella, and Enterobacter. This widespread distribution likely results from horizontal transfer, as plasmids can disseminate via conjugation or mobilization. Consistently, approximately 74% of PDF-encoding plasmids were predicted to be mobile (Fig. 1c-d). Additionally, at least five plasmids have already been identified as phage-plasmids, which are also capable of horizontal transfer between bacteria (Pfeifer et al. 2022; Ares-Arroyo et al. 2024; Pfeifer and Rocha 2024).

Integron

The integron system is a powerful genetic mechanism that enables bacteria to rapidly adapt to changing environments by capturing, stockpiling, and rearranging gene-encoding cassettes (Escudero et al. 2015). These cassettes often contain genes crucial for bacterial adaptation, such as those conferring antibiotic resistance or phage-defense (Darracq et al. 2025; Kieffer et al. 2025). A promoter, known as the Pc promoter, is located at the beginning of the integron. Typically, cassettes are promoter-less, and their expression relies on their position within the integron. Thus, cassettes positioned near the beginning (attI site) are actively expressed, establishing a gradient of expression that decreases with increasing distance from the Pc promoter. Cassettes located far from the promoter are minimally expressed and essentially serve as a low-cost molecular reservoir of functions (Krin et al. 2023). However, distance from the Pc promoter is not the only determinant of cassette expression; other factors, such as promoter strength (Jové et al. 2010) and the translation rate of upstream cassettes (Jacquier et al. 2009; Carvalho et al. 2024), can also play a role.

We probed our integron cassette collection (see materials and methods) and identified 24 PDF cassettes across 20 bacterial strains, predominantly within the genus Vibrio (Figure S2b—Table S3). Notably, two integrons contain multiple PDF cassettes: V. anguillarum MHK3 harbors three PDF cassettes and V. kanaloae R17 harbors two. In V. metschnikovii 07_2421, two cassette arrays were found, each containing a PDF cassette. The phylogenetic analysis of integron-encoded PDF showed that all Vibrio PDF cassettes share a common origin, likely driven by horizontal transfer. In contrast, the PDF cassette identified in P. suwonensis is phylogenetically distinct, and sequence analysis revealed that its closest homolog is a PDF from an Alphaproteobacteria of the genus Chelatococcus, with which it shares 51% identity, supporting a separate evolutionary origin (Fig. 1e).

PDF cassettes are present in both complete integrons and CALIN (Cluster of attC site lacking integron integrase (Cury et al. 2016)), but all these strains have an integrase capable of mobilizing cassettes from CALINs. The short distance between the recombination site and the start codon (25 to 28 bp) likely precludes the presence of an internal promoter within the cassette, making PDF expression likely dependent on its position in the integron. Their positions in the integrons vary, with some located far from the integration site (attI), implying ancient acquisition, while others are close to the integration site, suggesting recent acquisition or remobilization, as is the case for the V. anguillarum str. S3-4/9 PDF cassette (8th position) (Fig. 1e). Additionally, in the CALINs of Vibrio sp. J502 and V. anguillarum VIB4 and J360, PDF cassettes were found in the first position, downstream of a putative promoter sequence which could lead to their expression as already described by Loot and colleagues (Loot et al. 2024).

Most PDF cassettes are located in sedentary chromosomal integrons (SCIs). However, one cassette is found in a mobile integron on a plasmid, in V. campbellii DS40M4 and the CALIN identified in P. suwonensis resembles a mobile integron that has been integrated in the chromosome, as it is characterized by heterogeneous attC sites (60 to 112 bp) and by the presence of the aminoglycoside resistance gene aadA1, typically found in mobile integrons, in its first position (Cury et al. 2016).

In summary, our results highlight the widespread presence of multiple PDFs in bacterial genomes. Most of these PDFs contain the essential catalytic amino acids, suggesting that they are likely active. In addition, the observation that PDFs are frequently associated with mobile genetic elements capable of horizontal dissemination reinforces the idea that they are likely to play an adaptive role.

To further investigate the possible adaptive function of accessory PDFs, we focused on Vibrio cholerae, a well-studied pathogen and model organism known to possess two PDFs (Margolis et al. 2000). Furthermore, given that integron cassettes carrying PDF are predominantly found in Vibrio species, V. cholerae offers an ideal system to explore the adaptive roles of accessory PDFs.

Vibrionaceae PDF Diversity

We examined the PDF diversity in Vibrionaceae using 136 complete genomes from RefSeq (Table S4). Nearly half of these species (46.7%) encode multiple PDFs, and 12 species—mostly from the Photobacterium genus—harbor three PDFs (8.8%) (Figure S3).

Phylogenetic analysis of PDFs from these genomes, together with 23 integron-associated PDFs absent from the reference set, revealed several distinct subfamilies (Fig. 2). All species possess a conserved canonical type 1b PDF, designated Def1, which is derived from a common ancestor and consistently found in an operon with the formyltransferase FMT, as previously reported in other bacteria (Meinnel and Blanquet 1993; Mazel et al. 1997).

Fig. 2.

Phylogenetic tree of integron-encoded and Vibrionaceae PDF. The phylogenetic tree represents PDFs from 136 Vibrionaceae genomes (listed in Table S4) and 24 PDFs from integron cassettes (listed in Table S3). The colored outer ring indicates the PDF clades. UFBoot values over 80 are shown in a yellow-red gradient. The Tree was rooted using tree Def sequences from Gram-positive bacteria as an outgroup.

A second major clade, designated Def2, comprises PDFs found in Vibrio integron cassettes as well as most accessory PDFs in Vibrionaceae, including the type 1b variant previously described in V. cholerae Def2_VCH (Margolis et al. 2000). This group, present in 41.6% of Vibrionaceae (Figure S3), exhibits greater sequence variability than Def1, suggesting a potential adaptive role. Motif analysis further revealed that 61.8% of Def2 variants carry a mutation in motif I, where the second alanine is replaced by serine—without affecting enzymatic activity (see below).

To investigate the evolutionary origin of Def2 PDFs, we compared their phylogeny with that of the host species. The lack of congruence between the two trees suggests that accessory Def2 PDFs were disseminated by horizontal gene transfer, most notably between “cholerae” and “vulnificus” lineages (Figures S3 and S4).

We also found that the PDFs previously described in V. tubiashii and V. penaeicida cluster within the Def2 group (Fig. 2). These PDFs are located in a biosynthetic gene cluster (BGC) comprising two non-ribosomal peptide synthetases (NRPS), a type 1 polyketide synthase (PKS), and a methyltransferase (Cordero et al. 2012; Costa et al. 2024). This BGC has been proposed to produce a PDF inhibitor, with the cluster-encoded PDF acting as an antidote (Costa et al. 2024). Recent studies confirmed this model, showing that the metabolite indeed blocks PDF activity by binding its active site (Rill et al. 2025). PDF inhibition was further shown to trigger prophage activation in Vibrio through a non-SOS pathway (Chen et al. 2025).

Finally, beyond Def1 and Def2, we detected additional PDFs that are phylogenetically distinct from both major clades (Fig. 2). Since integron-associated PDFs belong to the Def2-like group, we focused our subsequent analyses on accessory Def2 proteins in V. cholerae.

Vibrio cholerae Harbor 2 Actives PDF

V. cholerae possesses two PDFs, Def1_VCH and Def2_VCH. The canonical PDF, Def1_VCH, is located on chromosome 1 within an operon that includes fmt, rsmB, trkA (as in E. coli (Cho et al. 2009)), trkH, and a gene of unknown function with a conserved DUF3157 domain (VC0041). The rsmB gene encodes a methyltransferase that specifically methylates cytosine 967 (^m5C967) of the 16S rRNA, stabilizing the binding of fMet-tRNA^fMet to the 30S pre-initiation complex before starting codon recognition (Burakovsky et al. 2012). The trkA and trkH genes are involved in potassium transport regulation (Zhang et al. 2020). In contrast, the accessory PDF, Def2_VCH, is encoded on chromosome 2 as a monocistronic unit. According to previously published proteomic data from the Vibrio cholerae N16961 WT strain grown in MH medium at 37 °C, the PDF Def1_VCH is 2 times more abundant than Def2_VCH (Fruchard et al. 2025).

Despite sharing only 51% amino acid sequence identity, Def1_VCH and Def2_VCH exhibit significant structural homology, with their most notable differences located in their C-terminal regions (Figure S5a). Amino acid sequence alignment reveals strict conservation of the three motifs that have been described as critical for catalytic activity. However, Def2_VCH shows a specific pattern variation in the amino acids, described in E. coli, as responsible for ribosome binding (Figure S5b). Specifically, the C-terminal helix of the E. coli PDF has been shown to insert into a groove between ribosomal proteins uL22 and uL32 (Bingel-Erlenmeyer et al. 2008; Sandikci et al. 2013; Akbar et al. 2021). The first half of this helix interacts with uL22 through residues L149, K150, R153, and K157, while the second half interacts with 23S rRNA through K160 and R163 (Bingel-Erlenmeyer et al. 2008; Akbar et al. 2021). In Def1_VCH, all amino acids crucial for ribosome binding are identical to E. coli PDF, while in Def2_VCH, R153 is substituted by a methionine and R163 by a lysine (Figure S5b). In E. coli, R153 forms a hydrogen bond with E52 of ribosomal protein uL22, and a methionine substitution likely disrupts this interaction. Interestingly, all PDFs within the Def2 group lack R153 (Figure S5c), instead, they have a methionine in 95% of cases, and more rarely an isoleucine or leucine, all hydrophobic amino acids (Figure S5d). Meanwhile, the glutamate E52 is conserved in 95% of Vibrionaceae uL22 proteins, the others have an aspartate (4%) or valine (1%) at this position. This suggests a possible functional divergence in their interaction with the ribosome.

We successfully deleted either def1_VCH or def2_VCH, suggesting that both PDFs are active. To further assess their essentiality, we attempted to construct a double mutant lacking both genes. Since PDF activity is known to be essential in many bacteria, we hypothesized that such a mutant would not be viable. To circumvent this, we maintained deformylase activity in trans using a thermosensitive plasmid to construct the double deletion mutant. To assess whether each PDF is functional, we introduced a compatible plasmid expressing either def1_VCH or def2_VCH and shifted the temperature to 42°C to cure the thermosensitive plasmid. Growth at 42°C was only observed when the second plasmid carried an active PDF, confirming that both Def1_VCH and Def2_VCH are enzymatically functional and can individually support bacterial viability (Fig. 3a).

Fig. 3.

Def2 confers resistance to ACT. a) In vivo PDF activity assay. Drop test of V. cholerae Δdef1_VCH Δdef2_VCH harboring a pSC101::def2_VCH-TS plasmid and a pSEVA228 plasmid at permissive temperature (left panel) or restrictive temperature (right panel). Growth at a restrictive temperature indicates that the PDF coding in the pSEVA228 plasmid is enzymatically active. b) Antibiotic susceptibility matrix of ACT. CMI was determined by broth microdilution as the minimal concentration of ACT with no observed growth. c) Antibiotic susceptibility matrix of LBM415. CMI was determined by broth microdilution as the minimal concentration of LBM415 with no observed growth. d to e) Structural superposition of Def1_VCH (blue) and Def2_VCH (green) in complex to ACT (Def1_VCH -ACT: red, Def2_VCH -ACT: orange). Side chains of amino acids involved in ACT and/or metal cofactor binding are shown in stick representation. f) Impact of the amino acid substitution on the resistance level to ACT. Antibiotic susceptibility matrix of ACT. CMI was determined by broth microdilution as the minimal concentration of ACT with no observed growth.

Def2 Reduces Susceptibility to ACT

Actinonin, the first discovered PDF inhibitor, is a natural antibiotic synthesized by a BGC consisting of non-ribosomal peptide synthetases (NRPS) and polyketide synthases (PKS), identified in Streptomyces species. It is plausible that ACT, along with other yet-to-be-identified natural PDF inhibitors, is present in the environment, where it may influence microbial competition. Integrons, which play a pivotal role in bacterial adaptation to environmental pressures, frequently harbor genes linked to antibiotic resistance and phage-defense (Darracq et al. 2025; Kieffer et al. 2025). Homologs of Def2_VCH are found within NRPS and PKS operons in some Vibrio species (Fig. 2) and in both chromosomal and mobile integrons (Fig. 2). These observations led us to hypothesize that accessory PDFs, such as Def2_VCH, might contribute to resistance against PDF inhibitors.

To test this hypothesis, we investigated whether the presence of the accessory PDF Def2_VCH provides an advantage in resistance to ACT. Deletion of the canonical PDF def1_VCH did not impact the sensitivity to ACT, but the deletion of the accessory PDF def2_VCH led to a 20-fold decrease in the MIC of ACT. Ectopic complementation of the Δdef2_VCH strain with a def2_VCH gene integrated in the chromosome restored sensitivity to levels observed in the wild-type (WT) and Δdef1_VCH strains, while complementation with the def1_VCH gene resulted in a more moderate gain in resistance (Fig. 3b). We further assessed resistance to another peptidomimetic PDF inhibitor, LBM415, which is derived from ACT and engineered by Novartis (Figure S6) (Anderegg et al. 2003; Fritsche et al. 2005; Rolan et al. 2011). In contrast, deletion of either def1_VCH or def2_VCH led to a comparable decrease in LBM415 resistance. Complementation of the Δdef2_VCH mutant with def1_VCH conferred only a slight increase in resistance compared to complementation with def2_VCH, suggesting a more limited protective effect of the canonical PDF against this inhibitor (Fig. 3c).

Since resistance depends on Def2_VCH despite its lower abundance, dosage effects can likely be ruled out. We therefore propose that the key factor is differential binding affinity to ACT. To explore this possibility at the molecular level, we determined high-resolution crystal structures of both proteins in complex with ACT (Fig. 3d-e and Table S5). In both structures, ACT binds in the same location within the catalytic pocket and adopts a nearly identical orientation and conformation, blocking access to the active site located at the bottom of a deep cavity in the protein core (Figure S7a-b).

In Def1_VCH, all previously identified amino acids involved in ACT binding in E. coli PDF were also identified interacting with ACT (Fig. 3d-e and Figure S5) (Clements et al. 2001). In contrast, Def2_VCH exhibited a different binding pattern. This is mainly due to the substitution of arginine 97 with a tyrosine, which weakens the interaction with ACT by abolishing important hydrogen bonds between the NH1 and NH2 atoms of Arg97 and solvent-exposed oxygen atoms of ACT (namely O20 and O27). Additionally, in the hydrophobic pocket containing the ACT P1' aliphatic chain, isoleucine at position 86 and leucine at position 125 in Def1_VCH are substituted by glycine and phenylalanine, respectively, in Def2_VCH (Fig. 3d-e and Figure S5b). These substitutions likely contribute to the variation in ACT binding affinity between Def1_VCH and Def2_VCH, explaining their different capacities to confer resistance to this inhibitor.

To further investigate the role of these amino acid substitutions in ACT resistance, we engineered reciprocal mutations, swapping the residues implicated in ACT interaction between Def1_VCH and Def2_VCH. Substitution of arginine 97 with tyrosine in Def1_VCH resulted in a significant increase in the MIC to ACT, reaching levels comparable to those conferred by Def2_VCH (Fig. 3f). Conversely, substitution of tyrosine 97 with arginine in Def2_VCH led to a marked reduction in the MIC, aligning with the resistance phenotype observed for Def1_VCH (Fig. 3f). Substitutions of glutamate 87 to glutamine (E87Q) and leucine 125 to phenylalanine (L125F) in Def1_VCH had either no or intermediate effects, in the same direction as the R97Y substitution, and their effects appeared to be cumulative (Fig. 3f). Those substitutions are conserved among the whole Def2 group and among BGC-encoded PDF identified in Vibrio species. It is therefore likely that these accessory PDFs, including the integron-encoded one, may also confer better resistance to actinonin or other PDF inhibitors in their native host.

These results indicate that the accessory PDF, Def2_VCH, plays a crucial role in resistance to actinonin in V. cholerae. The resistance conferred by Def2_VCH appears to be linked to a lower affinity of ACT for this protein compared to the canonical PDF, Def1_VCH. ACT is a natural product produced by a BGC in Streptomyces bacteria, suggesting that PDF inhibitors may be present in the environment and play a role in bacterial competition.

Integron PDF are Enzymatically Active

All integron-encoded Def2-like proteins exhibited the A48S substitution in motif I (Figure S5b), a feature also present in ∼62% of Def2 proteins overall. To assess its functional impact, we tested the enzymatic activity of three integron-encoded PDFs: VaDef from V. anguillarum str. MHK3, VkDef from V. kanaloae str. R17, and PsDef from Pseudoxanthomonas suwonensis str. 11-1. Using the previously described V. cholerae mutant lacking both endogenous PDFs, we showed that all three variants were functional and able to support cell viability (Fig. 4a). These results also demonstrate that the A48S substitution does not impair PDF activity in V. cholerae and expand the diversity of catalytic motifs compatible with enzymatic function.

Fig. 4.

Activity and resistance profile of integron PDF. a) In vivo PDF activity assay. Drop test of V. cholerae Δdef1_VCH Δdef2_VCH harboring a pSC101::def2_VCH-TS plasmid and a pSEVA228 plasmid at permissive temperature (left panel) or restrictive temperature (right panel). Growth at a restrictive temperature indicates that the PDF coding in the pSEVA228 plasmid is enzymatically active. VaDef, VkDef, and PsDef correspond to Vibrio anguillarum str. MHK3 integron PDF, Vibrio kanaloae str. R17 integron PDF and Pseudoxanthomonas suwonensis str. 11-1 integron PDF respectively. b) Antibiotic susceptibility matrix of ACT. CMI was determined by broth microdilution as the minimal concentration of ACT with no observed growth. c) Antibiotic susceptibility matrix of LBM415. CMI was determined by broth microdilution as the minimal concentration of LBM415 with no observed growth.

According to structural predictions from AlphaFold3 (Abramson et al. 2024), VaDef and VkDef, both belonging to the Def2 group, exhibit overall structure similarity with Def2_VCH (RMSD 0.38 Å over 151 alpha carbon and 0.39 Å over 153 alpha carbon, respectively). In contrast, PsDef, predicted to derive from a distinct evolutionary lineage (Fig. 1e), features an additional small C-terminal α-helix turn, typically absent in known type 1B PDFs (Figure S8a).

Integron PDF Cassette Confer Resistance to PDF Inhibitor

We assessed the resistance profiles of integron-encoded PDF cassettes VaDef, VkDef, and PsDef once integrated in single copy into the chromosome of the V. cholerae Δdef2_VCH mutant, under the control of the constitutive def2_VCH promoter. Unlike V. cholerae Def2_VCH, the Vibrio integron cassettes VaDef and VkDef conferred only limited resistance to ACT and LBM415 in V. cholerae (Fig. 4b-c), despite also having tyrosine at position 97. This limited resistance may be due to lower expression levels in V. cholerae, potentially resulting from differing codon usage. Consequently, these integron-encoded PDF cassettes might confer effective resistance to ACT in their native host.

In contrast, PsDef integration led to substantial resistance to actinonin, with at least a six-fold higher MIC compared to the WT strain and at least a 120-fold increase compared to the Δdef2_VCH mutant. Growth was observed even at the highest concentrations tested, preventing a precise MIC determination (Fig. 4b). PsDef also conferred strong resistance to LBM415, resulting in a 40-fold MIC increase relative to the WT strain and a 200-fold increase compared to either the Δdef1_VCH or Δdef2_VCH mutants (Fig. 4c). Furthermore, expressing PsDef in E. coli also conferred resistance to peptide deformylase inhibitors, with the ACT MIC increasing three-fold and the LBM415 MIC increasing 25-fold. (Figure S8c-d).

PsDef shares several characteristics with the Def2_VCH PDF of V. cholerae, notably the presence of tyrosine at position 97. It also features an insertion of a glycine between glutamates E87 and E88 of motif II (Figure S9b). Glutamate E87 likely participates in ACT binding as observed in E. coli PDF and V. cholerae Def1_VCH, while glutamate E88 is crucial for forming the hydrophobic pocket that accommodates the ACT P1' pentane chain (Clements et al. 2001). This amino acid insertion may alter the spatial positioning of E87, hindering its interaction with ACT. Such structural modifications could decrease ACT binding affinity, potentially explaining the notable resistance PsDef confers to peptide deformylase inhibitors.

A significant risk associated with the PsDef cassette is its potential capture by a class 1 mobile integron, which could facilitate its transfer across different bacterial species. To evaluate this risk, we assessed the recombinogenic capacity of the cassette and found it to be considerable (Figure S9a-b). Moreover, under the control of a native mobile integron Pc promoter, the expression of this cassette conferred significant resistance to both ACT and LBM415 in V. cholerae (Figure S9c-d).

Discussion

Despite the fact that the presence of accessory PDFs in certain bacterial genomes has been noted for over twenty years (Margolis et al. 2000; Guilloteau et al. 2002), no comprehensive study has yet addressed the global distribution and role of these accessory PDFs. We aimed at filling this gap and determined the distribution and genetic support of accessory PDFs and explored potential roles, which might explain their selection and maintenance in bacterial genomes.

Our large-scale analysis of complete genomes from RefSeq suggests that the formylation of initiator tRNA is likely conserved across most bacteria, as nearly all genomes encode a fmt gene. In addition, the presence of several PDF genes is a property common to ca. 50% of bacterial species. Additionally, we revealed significant intra-species variability for the presence of accessory PDFs, explained by their location on mobile genetic elements (MGEs) capable of horizontal gene transfer (HGT), such as plasmids and integrons. To our knowledge, PDF is the first gene involved in a conserved key translation process to be found in integron cassettes. However, a systematic scan of essential and core genes in integron cassettes has yet to be performed. The presence of the exact chromosomal copy of the E. coli PDF gene on plasmids in 12 Enterobacteriaceae species further demonstrated that PDFs could disseminate horizontally.

Until now, the role and activities of accessory PDFs had only been studied in a few species, notably S. aureus and S. pneumoniae, which possess an inactive type 3 accessory PDF (Margolis et al. 2000, 2001), and B. subtilis, whose accessory PDF was shown to have similar enzymatic activity to the canonical PDF (Haas et al. 2001). Our analysis of reference genomes coding a single PDF allowed us to redefine the three conserved motifs previously described (Giglione et al. 2004). This analysis revealed that most accessory PDFs retain the critical amino acids necessary for enzymatic activity. As 7% of bacterial species do not encode any PDF with all three motifs, this suggests that the precise presence of these motifs is not required for catalytic activity in all bacterial species. Additionally, earlier studies have shown that type 3 PDFs of T. brucei, with mutations in these motifs, are active (Bouzaidi-Tiali et al. 2007), indicating that even the accessory PDFs lacking full conservation of these conserved motifs might have deformylase activity in certain ecological contexts, as seen with many eco-paralogs (Mazel and Marlière 1989; Sanchez-Perez et al. 2008; Bratlie et al. 2010). Similar to many proteins in hyperhalophilic bacteria, which are active at high salinity levels but exhibit minimal or no activity under standard salinity conditions (Sanchez-Perez et al. 2008), accessory type 3 PDFs, which exhibit low or no deformylase activity in vitro, might become active in vivo under specific environmental stresses. In this line, given that PDFs are metalloproteases and that two of the conserved motifs are involved in binding divalent metals, it is plausible that type 3 PDF enzymes might adapt to oxidative stress conditions. During oxidative stress, divalent cations can be oxidized to trivalent forms, potentially inactivating standard metalloproteases. Type 3 PDFs might bind alternative cofactors, such as non-divalent metals, enabling deformylase activity under oxidative stress. This activity, although potentially lower than that of canonical PDFs, could be sufficient to support bacterial survival and stress response.

Our comparative genomics analyses showed that many Vibrionaceae species encode multiple PDFs, and most PDFs identified in integron cassettes are found in Vibrio genomes. Furthermore, the presence of two PDFs in V. cholerae was identified 20 years ago, but their roles and activities have not been studied yet. Our study revealed that about half of the Vibrionaceae species encode an accessory PDF, Def2, which belongs to the same phylogenetic group as PDFs found in Vibrio integron cassettes. Def2-like PDFs have strong structural homology with canonical Def1 PDFs but diverge at the C-terminal tail, involved in ribosome association. As the ribosomal exit tunnel is particularly dynamic, with many co-translational proteins interacting near or overlapping the PDF binding site (Denks et al. 2017; Bhakta et al. 2019; Akbar et al. 2021; Bögeholz et al. 2021), an accessory PDF with different ribosome association dynamics could alter co-translational protein binding on the ribosome, affecting nascent peptide processing. Moreover, the formyl group on the initiator methionine was recognized as an N-terminal degradation signal (N-degron), regulating nascent protein quality control (Piatkov et al. 2015). Having multiple PDFs might enhance the flexibility process, especially under stress conditions that affect ribosomal structure.

In V. cholerae, we showed that both PDFs, Def1_VCH and Def2_VCH, are active and support bacterial growth in rich media. Accessory Def2_VCH confers greater resistance to ACT, a natural PDF inhibitor produced by Streptomyces bacteria, compared to the canonical Def1_VCH. Characterization of both V. cholerae PDFs complexed with ACT revealed that Def2_VCH exhibits a lower affinity for ACT than Def1_VCH, attributed to amino acid substitutions affecting ACT binding, thereby explaining the increased resistance observed. Given the phylogenetic similarity of integron PDF cassettes in Vibrio species to the Def2 group, we hypothesized that these cassettes might also confer resistance to ACT in V. cholerae. However, the Vibrio integron PDF cassettes tested conferred lower resistance than Def2_VCH, despite sharing similar substitutions affecting ACT affinity, likely due to suboptimal codon usage.

It is unlikely that the presence of actinonin in the environment has driven the selection of accessory PDFs in Vibrio, given that Streptomyces and Vibrio do not occupy overlapping ecological niches. Instead, it is more plausible that other, yet unidentified, biosynthetic gene clusters (BGCs) producing PDF inhibitors exist within Vibrio's natural habitat. One such BGC, previously identified in Vibrio species, has been shown to mediate bacterial competition in marine environments (Cordero et al. 2012; Costa et al. 2024), and recent studies confirm that it produces a peptide-based PDF inhibitor (Chen et al. 2025; Rill et al. 2025). The presence of this BGC may therefore have contributed to the selection and maintenance of accessory PDFs in Vibrio genomes. Supporting this idea, another BGC producing a PDF inhibitor has also been identified in Rhodococcus fascians, with its co-encoded PDF conferring resistance to the compound (Ford et al. 2025). These findings suggest that naturally occurring PDF inhibitors are more widespread than previously thought, and that selective pressure imposed by such compounds may be a general driver for the acquisition of accessory PDFs in bacterial genomes, especially in integron cassettes. As antibiotic and phage resistance genes are frequently found in integrons (Darracq et al. 2025; Kieffer et al. 2025), it is possible that the capture of the PDF cassette was similarly favored by exposure to natural PDF inhibitors, while its persistence may be facilitated by the remarkable stability of integrons (Richard et al. 2024).

Finally, the integron PDF cassette from P. suwonensis (PsDef) has been shown to confer high resistance to both ACT and LBM415, even when expressed under a native mobile integron promoter. Furthermore, this cassette's high recombinogenic capability suggests it could be horizontally transmitted between bacterial species via plasmid-harboring mobile integrons. Due to its resistance phenotype, the dissemination of this cassette could pose a challenge for the therapeutic use of PDF inhibitors and question their viability in clinical applications.

Given that PDF inhibitors are likely present in bacterial environments, they probably contribute to bacterial competition. This environmental presence may have been selected for PDFs resistant to these inhibitors, partly explaining the widespread occurrence of multiple PDF genes in bacterial genomes. However, the exact mechanisms driving the selection and maintenance of accessory PDF genes are likely multifaceted, involving resistance to natural inhibitors and other stresses, such as those influencing ribosome integrity.

Materials and Methods

Bacterial Strains, Plasmids, and Primers

All bacterial strains and plasmids in this study are summarized in Table S6.

Bacterial Growth Conditions

Bacteria were grown in LB broth (Lennox) or Mueller-Hinton (MH) broth. Antibiotics were used at the following concentrations: carbenicillin (100 μg/mL), chloramphenicol (25 μg/mL for E. coli, 5 μg/mL for V. cholerae), kanamycin (25 μg/mL), and spectinomycin (50 μg/mL for E. coli, 100 μg/mL for V. cholerae). Diaminopimelic acid (DAP) was added at 0.3 mM, 5-bromo-4-chloro-3-indolyl-beta-D-galactopyranoside (X-Gal) at 40 µg/mL, glucose at 1% (w/v), arabinose at 0.2% (w/v), and IPTG at 250 μM.

PCR and Sequencing

For PCR cloning, Phusion High-Fidelity DNA Polymerase (High-Fidelity Phusion Master Mix, Thermo Fisher #F548S) was used. PCRs were performed on genomic DNA preparations using the DNeasy® Tissue Kit (Qiagen) or on plasmid preparations using the GeneJET Plasmid Miniprep Kit (Thermo Scientific #K0503). PCR products were purified using the GeneJET PCR Purification Kit (Thermo Scientific #K0701).

PCRs for cloning verification were conducted using DreamTaq Polymerase (Thermo Scientific #EP0705). These PCRs were performed directly on bacterial biomass from isolated colonies.

Sequencing PCR products was carried out by Eurofins, using their Sanger sequencing TubeSeq service.

Plasmid Construction

The plasmids used in this study are listed in Table S6. Briefly, plasmids with an oriV(SC101) origin of replication were derived from pMP400, with pBAD promoter and araC removed. Plasmids with a p15A origin were derived from pSU38, with lacZα removed and KmR cassette replaced by ampR cassette (Bartolomé et al. 1991). Plasmids with an oriV(RK2) origin of replication were derived from pSEVA228 (Calero et al. 2016), with xylS removed. Plasmids for Tn7 transposition were derived from pMP234 (de Lemos Martins et al. 2018), and protein expression plasmids were derived from pET22b + with the pelB signal sequence for periplasmic secretion removed (Novagen #69744).

Plasmid constructions were performed using Gibson Assembly (Gibson et al. 2009) for complex cloning and inverse PCR, followed by ligation for point mutations. Gibson Assembly was carried out as described by (Rabe and Cepko 2020). Inverse PCR was performed with Phusion High-Fidelity DNA Polymerase (Thermo Fisher #F548S), and the resulting products were ligated using T4 DNA Ligase (Thermo Scientific #EL0011).

Constructed plasmids were transformed into E. coli cloning strain NGEpir containing the RK2 plasmid integrated into its genome, allowing the transfer of the generated plasmids via bacterial conjugation. Transformations were performed by electroporation at 1.8 kV/cm, and transformants were selected on MH agar plates supplemented with the appropriate antibiotics and substrate.

For plasmids containing peptide deformylase genes within integron cassettes (pMLB42 to pMLB44), the entire cassette was synthesized as gBlocks Gene Fragments (IDT) and cloned into the pCR-Blunt II-TOPO vector using the Invitrogen Zero Blunt™ TOPO™ PCR Cloning Kit (#450245). These constructs were transformed into E. coli cloning strain π3,813.

For the plasmid containing the def-fmt fusion of Exiguobacterium alkaliphilum, the entire gene was synthesized as gBlock Gene Fragments (IDT) with optimized codon usage for E. coli MG1655.

Quality controls of all plasmid constructs were done by Sanger sequencing and restriction digestion analysis.

Bacterial Conjugation

Plasmid transfer was performed by bacterial conjugation using an E. coli auxotrophic for diaminopimelic acid (DAP) and containing an integrated RK2 plasmid in the chromosome (E. coli NGEpir strain), which allows the mobilization of plasmids with an RK2/RP4 origin of transfer (oriT). The donor strain also carries the pir gene for the replication of plasmids with an R6K origin.

Conjugation was conducted on Mueller–Hinton (MH) agar plates supplemented with DAP for at least 3 h. After mating, transconjugants were selected on appropriate antibiotic plates without DAP to eliminate the donor strain.

Creation of def1_VCH and def2_VCH Deletion Mutants

Deletion mutants of def1_VCH (VC_0046) and def2_VCH (VCA_0150) were generated in Vibrio cholerae serotype O1 biotype El Tor strain N16961 hapR + using natural co-transformation, as described by de Lemos Martins et al. 2018. Briefly, V. cholerae cells were rendered competent by growth on chitin, following the protocol by Meibom et al. 2005. Once competent, a small amount of PCR product containing an antibiotic resistance cassette flanked by 6 kb of homologous DNA from the intergenic region VC1903/1902, along with an excess of PCR product containing the desired deletion fragment, was added to the medium. This method relies on the principle that if a cell successfully integrates the resistance cassette, it is likely to have also been transformed by the excess deletion fragment (Dalia et al. 2014).

To delete def1_VCH, 100 ng of a spectinomycin resistance cassette (aadA7) was co-transformed with 1 µg of a DNA fragment containing the desired deletion. Specifically, the homologous regions consisted of 3 kb upstream of the ATG start codon and 3 kb downstream of the stop codon of def1_VCH, ensuring a total of 6 kb of flanking sequences. Transformants were selected on spectinomycin plates, and correct genome editing was confirmed by PCR and sequencing of the resulting loci. Approximately 50% of the bacteria that integrated the resistance cassette also incorporated the deletion.

The def2_VCH deletion was performed similarly, but with the presence of a thermosensitive pSC101 plasmid expressing def2_VCH under a strong constitutive promoter (pMLB53-TS). The co-transformation and selection were carried out at 30 °C, and the media were supplemented with 50 µg/mL of carbenicillin to maintain the expression plasmid. As for the def2_VCH deletion mutant, transformants were selected on spectinomycin plates, and correct genome editing was confirmed by PCR and sequencing of the resulting loci. Approximately 50% of the bacteria that integrated the resistance cassette also incorporated the deletion. After def2_VCH was deleted, the thermosensitive plasmid was cured by shifting the temperature from 30 °C to 42 °C.

For the double deletion mutant (def1_VCH and def2_VCH), def1_VCH was first deleted using the aadA7 cassette. Following this, def2_VCH was deleted using the strategy described for the single def2_VCH deletion, but this time using a chloramphenicol resistance cassette (cat) to replace the spectinomycin resistance cassette at the same integration site. The thermosensitive plasmid pMLB53-TS was kept in the double mutant to ensure the presence of at least one functional PDF, which is essential for bacterial growth, by growing the mutants at a permissive temperature.

Tn7-Insertion

For ectopic complementation with PDF genes of the different mutants, we inserted a copy of the gene into the att-Tn7 site present on the chromosomes of either E. coli or V. cholerae, near the glmS gene, following the strategy described in (de Lemos Martins et al. 2018). The helper plasmid pMVM1 was first transformed into both V. cholerae N16961 and E. coli MG1655. This plasmid has a thermo-sensitive pSC101 origin of replication and carries a pBAD promoter that induces the expression of TnsABCD transposases, which catalyze insertion into att-Tn7 at high frequency as described by McKenzie and Craig 2006.

A second shuttle vector derived from pMP234 was used for the specific integration of the different PDF genes (Fournes et al. 2021). This vector contains the Tn7 IR sequences recognized by the transposases and was modified to carry the specific PDF gene along with an aph cassette for kanamycin resistance, which allows selection for transposition events. The pMP234 vector is a conditionally replicative vector, which cannot replicate in recipient cells that lack the Π protein.

For transposition, the pMP234 shuttle vector was delivered by conjugation into the recipient strains containing pMVM1. Transposition events were selected by plating conjugants on kanamycin plates without DAP. These plates were incubated overnight at 42°C to eliminate the helper vector.

All the PDF genes integrated in the chromosome using this set-up are controlled by the same promoter, that of the V. cholerae def2_VCH gene.

Minimum Inhibitory Concentration (MIC) Determination

For ACT (Sigma-Aldrich #A6671) and LBM415, MICs were determined by broth microdilution according to EUCAST guidelines, as summarized by (Kowalska-Krochmal and Dudek-Wicher 2021). First ACT and LBM415 were diluted in DMSO to concentrations of 26 and 23 mM, respectively. Bacteria overnight cultures were adjusted to an OD600 of 1 and diluted 1:1,000 in MH broth (approximately 5 × 10⁵ CFU/mL). These dilutions were placed in Falcon® 96-well Clear Flat Bottom TC-treated Culture Microplates with Lids (#353072), and the corresponding antibiotic concentrations were added. For control wells without antibiotics, DMSO was added to match the highest volume used in the antibiotic wells to ensure that DMSO itself did not have a toxic effect. The plates were incubated at 37°C with agitation in a TECAN® infinite M200 Pro plate reader for 24 h, and the absorbance at 600 nm was measured every 15 min. The MIC was defined as the lowest concentration of antibiotic at which no bacterial growth was observed. Each experiment was performed in biological triplicate with at least three independent repetitions.

In Vivo PDF Activity Assay in V. cholerae

To assess the in vivo activity of peptide deformylase (PDF) in V. cholerae, a double mutant strain (Δdef1_VCH Δdef2_VCH) harboring the plasmid pMLB53-TS, which expresses def2_VCH in trans under the expression of a constitutive P_trc promoter, was used. A second plasmid with a compatible oriV-trfA origin of replication (backbone pSEVA228) encoding the PDF of interest, under the expression of the constitutive P_def2VCH promoter, was introduced via conjugation. The bacteria containing both plasmids were grown overnight at 30 °C with appropriate antibiotics.

Overnight cultures were serially diluted, and 10 µL of each dilution was spotted onto MH agar plates supplemented with kanamycin to maintain the presence of the pSEVA228 plasmid. Plates were incubated at 30 °C and 42 °C. Growth at 42 °C indicated active PDF expression from the pSEVA228 plasmid, as bacteria with inactive PDF would be unable to grow at the higher temperature.

Computational Genomic Analyses

Databases Used

To conduct a comprehensive analysis of peptide deformylases (PDFs) across various bacterial species, multiple genomic databases were employed. For species-wide analysis, reference genomes from RefSeq were utilized, encompassing a total of 5,042 genomes as of May 4, 2024.

To examine intra-species variability, complete genomes available in RefSeq were analyzed, consisting of 40,298 genomes as of April 28, 2024.

Plasmid-borne PDFs were identified using the PLSDB, which included 50,554 replicons as of November 23, 2023. Replicons larger than 1 Mb were excluded to minimize the inclusion of secondary chromosomes and chromids (see Methods for details on this filtering rationale).

Additionally, the presence of PDFs within integron cassettes was investigated. For this, we detected integrons in bacterial complete genomes from RefSeq (32,798 genomes as of May 15, 2023) using IntegronFinder 2.0.5 (Néron et al. 2022). We used default parameters, with –local-max for accurate detection and –calin-threshold 1 to also detect SALINs (Single attC site lacking integron-integrase). For each genome, integrons were detected replicon by replicon. IntegronFinder provides three subtypes of integron-like elements: complete integrons (IntI plus a few cassettes flanked by attC sites), CALINs (Clusters of attC sites lacking an integrase), and loner integrases. A CALIN with only one attC is classified as a SALIN. In the current study, we considered every cassette protein belonging to integrons or CALINs detected by IntegronFinder. This procedure yielded a collection of 115,707 cassettes containing coding sequences, which we then used to identify integron-encoded PDFs. The collection of integron cassettes is publicly accessible at https://zenodo.org/records/17436480. For assistance with reuse or questions regarding the dataset, please contact Didier Mazel (didier.mazel@pasteur.fr) or Eduardo Rocha (eduardo.rocha@pasteur.fr).

For a detailed analysis of PDFs within the Vibrionaceae family, genomes of 136 species with complete sequences available in RefSeq as of January 3, 2023, were examined. The exact list of these genomes is provided in Table S4.

PDF Identification and Annotation Verification

Because no standardized method exists for detecting peptide deformylases (PDFs), we applied two complementary strategies to detect PDF genes in complete bacterial reference genomes from RefSeq (5,042 genomes). The first relied on annotations from the NCBI Prokaryotic Genome Annotation Pipeline (PGAP) (hereafter referred to as the PGAP method), while the second used an HMM search with the Pfam profile PF01327 after re-annotation of all genomes with Prokka (v.1.14.5, default parameters) (hereafter referred to as the HMM method).

We then compared the results obtained with the two approaches. In 4,966 genomes (98.5%), both methods returned identical PDF copy numbers. In the remaining 76 genomes, the results differ depending on the method. In 8 cases, no hit was detected by the HMM method, whereas PGAP annotated a PDF. In 30 cases, both methods detected multiple PDFs, but the copy number varied (e.g., two versus three copies). Finally, in only 38 genomes (0.75%), the classification differed between “single” and “multiple” PDFs depending on the method.

Given this marginal discrepancy, we opted to rely on the PGAP method for the detection of PDF in all complete RefSeq genomes, since it provides standardized annotation while avoiding the substantial computational cost of re-annotating 40,000 genomes.

For the detection of PDF genes in plasmid sequences, all 50,554 replicons from the PLSDB database were annotated using Prokka (v.1.14.5), followed by homology searches with HMMER against the same Pfam profile (PF01327). Similarly, the PDF gene in the integron cassette collection was detected via HMMER searches with PF01327.

PDF Genomic Context Analysis

To investigate the genomic context of PDF genes, we extracted the five coding sequences (CDSs) immediately upstream and downstream of each PDF gene across the bacterial reference genomes. Gene annotation was performed using Prokka (v1.14.6).

For the identification of Methionyl-tRNA formyltransferase (fmt) genes, we retrieved all CDSs annotated as fmt by Prokka that were located directly downstream of PDF genes and encoded on the same DNA strand. These FMT protein sequences were aligned using MAFFT (−auto), and poorly aligned regions were trimmed with TrimAl. A hidden Markov model (HMM) profile was then constructed and used to search for FMT homologues using HMMER v3.3.2. Although the HMM profile also detects structurally related formyltransferases (e.g., PurN, PurU, ArnA), applying a stringent e-value cutoff of <10⁻²⁰ excludes these homologues and retains only bona fide FMT proteins. However, we note that the use of this strict e-value threshold may lead to the omission of highly divergent FMT sequences, particularly from phylogenetically distant taxa relative to those used to construct the HMM profile.

Plasmid Mobility Analysis

Plasmid mobility was assessed using MacSyFinder 2.1.3 with the CONJScan model (Guglielmini et al. 2014; Néron et al. 2023). Plasmids encoding PDFs were classified into mobility types as follows: (i) conjugative plasmids (pCONJ), which encode a complete mating pair formation (MPF) system along with a relaxase; (ii) decayed conjugative plasmids (pdCONJ), which encode an incomplete MPF system but retain a relaxase; and (iii) mobilizable plasmids (pMOB), which encode a relaxase but lack MPF components. Plasmids carrying only an oriT were identified using BLASTn (v2.15.0) as described by Ares-Arroyo et al. 2024. Mobilizable plasmids encoding either a relaxase or an oriT can be horizontally transferred by hijacking the conjugation machinery of co-resident conjugative elements.

Core Genome Analysis

The core genome of Vibrionaceae was determined using PanACoTA v1.3.1 with a sequence identity threshold of 65% (-i 0.65) and a presence criterion in at least 95% of the analyzed genomes (-t 0.95). The dataset included 136 complete Vibrionaceae genomes (Table S4), resulting in the identification of 735 persistent gene families. Each family contained exactly one representative in at least 130 of the 136 genomes (95.0%), while the remaining genomes lacked homologs. Core genome alignment was performed using MAFFT v7.467 with the –auto option. A maximum-likelihood phylogenetic tree was then constructed with IQ-TREE v2.0.6 (Minh et al. 2020) using 1,000 ultrafast bootstrap replicates (Hoang et al. 2018) and the -m TEST algorithm to determine the best-fit model, which was identified as GTR + F + I + G4 according to the Bayesian Information Criterion (BIC).

PDFs Phylogenetic Analysis

For the phylogenetic analysis of peptide deformylases (PDFs), amino acid sequences were similarly aligned using MAFFT (option –auto, v.7.467), or using MUSCLE super5 (v.5.1) to generate Multiple Sequence Alignments (MSAs) by permuting the guide tree (“none”, “abc”, “acb”, “bca”) (Katoh and Standley 2013; Edgar 2022). A phylogenetic tree was constructed with IQ-TREE (v.2.2.2.2). This time, 1,232 protein models were tested, and the best-fitting model was selected according to the BIC. The robustness of the tree was evaluated using 1,000 bootstrap replicates and the SH test replicates.

Pfold Determination

The folding probability of integron cassette attC sites, called Pfold, was determined as described in (Loot et al. 2017; Vit et al. 2020). Briefly, the Vienna RNA fold WebServers (version 2.6.3) were used with the option “avoid isolated base pair” unchecked, and the DNA parameters model (Matthews model, 2004) selected. Pfold was calculated using the formula: $pFold=e^{\frac{\mathrm{\Delta }Gu-\mathrm{\Delta }Gc}{RT}}$. Where ΔGu is the Gibbs energy unconstrained, ΔGc is the Gibbs energy constrained, R is the gas constant in kcal/mol/K (1.9120458 × 10⁻³), and T is the temperature at 37 °C in Kelvin (310.15 °C).

attC-attI Recombination Frequency

The attC-attI recombination frequency was determined as described in (Vit et al. 2020, 2021). Briefly, attC sites were cloned into a conditionally replicative (suicide) plasmid derived from pSW23T (oriR6K), which codes for the cat gene conferring chloramphenicol resistance. This plasmid, which cannot replicate in recipient cells lacking the Π protein, was maintained in the donor strain E. coli β2163, which encodes the pir gene and can transfer the plasmid via conjugation. Two different attC sites were tested: the classical one from the aadA7 cassette and the one from the integron cassette of P. suwonensis 11-1 coding for a PDF. Each site was cloned in both orientations so that either the top or the bottom strand of the attC site is delivered in the recipient strain.

The recipient strain is an E. coli MG1655 harboring two plasmids. One plasmid derived from pBAD43, containing the aadA7 gene and an arabinose-inducible promoter allowing the expression of an integrase intI1 gene (pL294) or empty (pL290). The second plasmid is derived from pSU38 and contains the aph gene and an attI1 site (p929), allowing recombination with the attC site from the suicide plasmid.

The suicide plasmid was transferred by conjugation to the recipient strain on MH agar supplemented with DAP and arabinose, allowing the expression of the integrase. The suicide plasmid cannot replicate in the recipient strain, so maintaining chloramphenicol resistance requires recombination between the attC site on the suicide plasmid and the attI site on plasmid p929, forming a chimeric plasmid. In the absence of integrase, this recombination is not possible.

Following conjugation, transconjugants were selected on MH agar supplemented with chloramphenicol, kanamycin, and spectinomycin. Recipient cells were selected on MH agar with kanamycin and spectinomycin. 16 transconjugants were tested by PCR to verify the correct attC-attI recombination using the primer SwBeg-MFD when the attC bottom strand is delivered and the primers SwBeg-MRV when the top strand is delivered.

Recombination frequencies were calculated by dividing the number of transconjugants by the total number of recipients. If some of the 16 PCR-tested transconjugants did not show attC-attI recombination, the frequencies were recalculated as described by Vit et al. 2020.

Protein Purification

Four distinct plasmids were constructed for the expression of peptide deformylases, each featuring an N-terminal hexahistidine tag (6xHis) followed by a 3xGS linker. The plasmids and their encoded proteins are detailed in Table S6.

Protein Expression

E. coli Bli5 (BL21 (DE3) + pDIA17) (Munier et al. 1992) cells transformed with the respective pET22b plasmid encoding peptide deformylase (pMLB109, pMLB110) were grown in 1 L of LB medium supplemented with chloramphenicol and ampicillin at 37 °C in a 5L Erlenmeyer flask until the optical density at 600 nm (OD600) reached 0.8. Protein expression was induced by adding IPTG to a final concentration of 250 µM, and the culture was incubated for an additional 3 h at 25 °C. The cells were harvested by centrifugation at 8,000 × g for 15 min at 4 °C, and the resulting cell pellet was stored at −20 °C for up to one month prior to lysis.

Cell Lysis

The frozen cell pellet was resuspended in 50 mL of lysis buffer (25 mM Tris-HCl, pH 7.8, and 100 mM NaCl) and lysed using a CellD cell disruptor at 1.4 Kbar. The lysate was clarified by centrifugation at 25,000 × g for 30 min at 4 °C.

Affinity Chromatography

The clarified lysate was loaded onto a HisTrap™ High Performance 5 mL column (Cytiva 17-5248-02) using an Äkta Start system. The column was equilibrated with lysis buffer (25 mM Tris-HCl, pH 7.8, and 100 mM NaCl) before sample application. Following the sample application, the column was washed with 10 column volumes of the same buffer at a flow rate of 2 mL/min. Protein elution was carried out with a gradient of buffer A (25 mM Tris-HCl, pH 7.8, 100 mM NaCl, and 20 mM imidazole) and buffer B (25 mM Tris-HCl, pH 7.8, 100 mM NaCl, 300 mM imidazole, and 10% glycerol). The gradient was run over 40 mL, increasing buffer B from 0% to 100% at a flow rate of 2 mL/min, and fractions were collected every 2 mL. The protein of interest started to elute at approximately 50% buffer B (150 mM imidazole) and was concentrated on Vivaspin® 6 Centrifugal Concentrator Polyethersulfone with a molecular weight cutoff of 10 KDa to a final volume of 5 mL.

Size Exclusion Chromatography

For further purification, the 5 mL protein solution from affinity column chromatography was subjected to size exclusion chromatography using a HiLoad 16/600 Superdex 75 pg column on an UPC-900 system. The column was equilibrated with lysis buffer (25 mM Tris-HCl, pH 7.8, and 100 mM NaCl), and the protein was eluted in the same buffer at a flow rate of 1 mL/min. Fractions containing the target protein were collected, pooled, and concentrated on Vivaspin® 6 Centrifugal Concentrator Polyethersulfone with a molecular weight cutoff of 10 KDa to a final concentration of 10 mg/mL for downstream applications.

SDS-PAGE Analysis

Fractions were analyzed by SDS-PAGE using Mini-PROTEAN TGX Stain-Free gels (4 to 15% acrylamide, Bio-Rad) in a running buffer of 1× Glycine-Tris-SDS (19.2 mM glycine, 2.5 mM Tris, and 0.01% SDS, pH 8.3). Electrophoresis was performed at 170 V, and the gel was stained with Coomassie Brilliant Blue R-250 Staining Solution (Bio-Rad #1610436) for 20 min, followed by destaining with Coomassie Brilliant Blue R-250 Destaining Solution (Bio-Rad #1610438) to assess protein purity.

Quality Control Analysis

The quality control of the recombinant proteins was assessed by mass spectrometry and dynamic light scattering (DLS) according to the ARBRE-MOBIEU/P4EU guidelines (de Marco et al. 2021).

Crystallization

Screening of crystallization conditions and optimization of hits were performed at the Crystallography Core Facility of the Institut Pasteur (Weber et al. 2019). Briefly, initial screenings were performed in 96-well Greiner plates using a Mosquito automated nanoliter dispensing system (TTP Labtech, Melbourn, UK). The plates were then stored at 18 °C in a RockImager automated imaging system (Formulatrix, Bedford, USA) to monitor crystal growth. Initial crystallization hits were optimized in 24-well plates using the hanging drop method. The best Def1_VCH crystals grew in wells containing 20% (w/v) PEG4K, 20% (v/v) 2-Propanol, and 0,1 M Na3-citrate, pH 5.6 in the reservoir. Def2_VCH crystals were obtained in 20% (w/v) PEG4K, 10% (v/v) 2-Propanol, 0,1 M Hepes, pH 7.5. The crystals were snap frozen in liquid nitrogen for data collection using the crystallization solution as cryoprotectant for Def1_VCH and a mixture of oils (50% paratone N + 50% paraffin) for Def2_VCH.

Diffraction Data Collection and Structure Determination

X-ray diffraction data were collected at beamlines PROXIMA 1 and PROXIMA 2a (Synchrotron SOLEIL, St. Aubin, France) and processed with autoPROC (Vonrhein et al. 2011). The crystal structures were solved by the molecular replacement (MR) method using Phaser (McCoy et al. 2007), and trimmed AlphaFold (Jumper et al. 2021; Mirdita et al. 2022) models as a search probe. The final models were obtained through interactive cycles of manual model building with Coot (Emsley and Cowtan 2004) and reciprocal space refinement with Buster (Bricogne et al. 2017) and Phenix (Liebschner et al. 2019). Figures were generated using ChimeraX.

Accession Codes

Atomic coordinates and structure factors have been deposited in the RCSB Protein Data Bank under the accession codes 9QFR (Def1_VCH-ACT) and 9QFT (Def2_VCH-ACT).

Supplementary material

Supplementary material is available at Molecular Biology and Evolution online.

Funding

Our laboratory is funded by the Institut Pasteur, France and the Centre National de la Recherche Scientifique. This work was supported by the Fondation pour la Recherche Médicale (EQU202103012569 and FDM202106013531, to M. Lambérioux). This work was also supported by Pfizer Innovation France.

Data Availability

All genomic datasets used in this study originate from publicly available databases such as NCBI, RefSeq, and PLSDB. All software and scripts used for analysis are publicly accessible on GitHub. The structural data have been deposited in the Protein Data Bank (PDB) under unique identifiers: 9QFR and 9QFT. The collection of integron cassettes is publicly accessible online at https://zenodo.org/records/17436480.