Arginine-rhamnosylation as new strategy to activate translation elongation factor P

Abstract

Ribosome stalling at polyproline stretches is common and fundamental. In bacteria, translation elongation factor P (EF-P) rescues such stalled ribosomes, but only when it is post-translationally activated. In Escherichia coli, activation of EF-P is achieved by (R)-β-lysinylation and hydroxylation of a conserved lysine. Here we have unveiled a markedly different modification strategy in which a conserved arginine of EF-P is rhamnosylated by a glycosyltransferase (EarP) using dTDP-l-rhamnose as a substrate. This is to our knowledge the first report of N-linked protein glycosylation on arginine in bacteria and the first example in which a glycosylated side chain of a translation elongation factor is essential for function. Arginine-rhamnosylation of EF-P also occurs in clinically relevant bacteria such as Pseudomonas aeruginosa. We demonstrate that the modification is needed to develop pathogenicity, making EarP and dTDP-l-rhamnose-biosynthesizing enzymes ideal targets for antibiotic development.

Ribosomes translate an mRNA sequence into a polypeptide chain. During this process, specific X-PP-X tripeptide sequence motifs can induce ribosome stalling^1–6. Eukaryotic/archaeal initiation factor 5A (e/aIF5A) and its bacterial ortholog, EF-P, alleviate the stalled ribosomes by binding and stimulating peptide bond formation^3,4,7–10. With its three β-barrel domains, the L-shaped EF-P is structurally reminiscent of transfer RNA (tRNA)¹¹ and binds to the ribosome between the sites of peptidyl-tRNA binding (P-site) and tRNA exiting (E-site)¹². A positively charged residue at the tip of the loop region in domain I of EF-P protrudes toward the peptidyl-transferase center and can reach into it when elongated by modification¹². Accordingly, the conserved lysine in e/aIF5A is extended to hypusine by deoxyhypusine synthase and deoxyhypusine hydroxylase^13–15. Analogously, in bacteria such as E. coli the protruding lysine (K34) of EF-P is (R)-β-lysinylated and hydroxylated by the concerted action of EF-P lysyl-transferase (EpmA), lysine amino-mutase (EpmB) and EF-P hydroxylase (EpmC; Fig. 1a)^16–20. Here we report the identification of an EF-P subfamily activated by a chemically different modification. Using Shewanella oneidensis as a model organism, we found rhamnosylation of a conserved arginine. We also identified the corresponding glycosyltransferase, EarP. This modification is not only crucial for bacterial fitness but also for pathogenicity in P. aeruginosa, and thus it might be equally important in other clinically relevant species such as Neisseria gonorrhoea or Bordetella pertussis.

Figure 1. Bioinformatic identification of the EarP-arginine type EF-P subfamily.

(a) (R)-β-lysinylation and hydroxylation of EF-P in E. coli. (b,c) A 31-amino-acid (aa)-long sequence logo of EF-P encompassing a part of domain I, including the loop region. Arrows point to the positively charged residue at the tip of the loop region. Numbering depicts aa in the polypeptide chain. (b) Weblogo generated from EF-P of bacteria encoding EpmA and EpmB. (c) Weblogo generated from EF-P of bacteria derived from the newly identified branch that co-occurs with EarP.

RESULTS

Identification of the EarP-arginine type EF-P subfamily

Whereas the hypusinylation pathway is highly conserved in archaea and eukaryotes, EpmA (also known as YjeA, PoxA, GenX) and EpmB (YjeK) are only found in about 26% of all bacteria²⁰ (Supplementary Results, Supplementary Figs. 1–5 and Supplementary Data Set 1). The third modification enzyme, EpmC (YfcM), co-occurs with EpmA and EpmB but is restricted almost exclusively to γ-proteobacterial genomes (Supplementary Figs. 4,5 and Supplementary Data Set 1) corroborating its minor role in EF-P function^3,6–8,21. We hypothesized that the genes encoding EF-P and the associated modification system have coevolved. In a phylogenetic analysis of EF-P sequences, we identified a distinct subfamily, encoded in genomes lacking EpmABC orthologs, that has a strictly conserved arginine (R32) in the position equivalent to K34 in E. coli (Fig. 1b,c, Supplementary Fig. 4 and Supplementary Data Set 1). The members of this subfamily represent about 9% of all EF-Ps, but the distribution deviates from the currently accepted species phylogeny. As the newly identified EF-P branch encompasses all β-proteobacteria, we hypothesize this subdivision as the phylogenetic origin, with subsequent horizontal transfer into several γ-proteobacterial orders (including Pseudomonadales, Aeromonadales and Alteromonadales) as well as some Fusobacteria, Planctomycetes and Spirochetes (Supplementary Data Set 1). We took advantage of the anomalous EF-P phylogeny by searching for putative EF-P modification enzymes associated with this subfamily via gene neighborhood and co-occurrence using STRING²². This led us to identify a protein with a conserved domain of unknown function (DUF 2331), which we designated as EarP. Its distribution strictly coincides with the newly identified EF-P subfamily (Supplementary Fig. 5). Moreover, the corresponding gene, earP, always lies within a four-gene distance to efp; in 94% of cases, both genes are directly adjacent (Supplementary Fig. 6a).

EF-P and EarP are functionally linked

To investigate whether EF-P and EarP are functionally linked, we used the ubiquitous, facultative anaerobic, alteromonadal γ-proteobacterium S. oneidensis. Bacteria of the genus Shewanella are commonly used in microbial fuel cells and have high potential in bioremediation because of their ability to use a wide range of terminal electron acceptors, including heavy metals^23,24.

In a first step, we generated markerless in-frame deletions of S. oneidensis efp (locus tag SO_2328) and earP (locus tag SO_2329) and phenotypically characterized the resulting mutant strains, Δefp_S.o. and ΔearP_S.o., respectively. Bacteria lacking efp, such as E. coli or Agrobacterium tumefaciens, have diminished growth rates^16,25. In line with these results, deleting either efp_S.o. or earP_S.o. increased the doubling time from 40 min to 110 min and 70 min, respectively, which was reversed by providing the corresponding gene copy in trans (Fig. 2a and Supplementary Fig. 6a). In parallel, we analyzed the growth of the Δefp_S.o. strain, which encodes an EF-P variant where the strictly conserved R32 (Fig. 1c) was substituted by either lysine (R32K) or alanine (R32A). Both strains phenocopied Δefp_S.o., demonstrating the importance of this conserved arginine for EF-P_S.o. function (Fig. 2a and Supplementary Fig. 6a). We also investigated whether the synthesis of polyproline-containing proteins is affected in the absence of efp_S.o. and earP_S.o.. Therefore, we used the reporter plasmid p3LC-TL30-3P, which encodes a LacZ variant that is preceded by a stretch of three proline residues (Fig. 2b)^3,7. As expected, the β-galactosidase activities of the Δefp_S.o. or ΔearP_S.o. strains were both reduced by about tenfold, providing clear experimental evidence that EarP is required for EF-P activity.

Figure 2. Phenotypic analysis of S. oneidensis MR-1 earP and efp deletion mutants.

(a) Growth of S. oneidensis MR-1 strains. Left, wild-type strains in comparison to Δefp_S.o. and ΔearP_S.o. deletion strains and after complementation in trans (+efp_S.o., +earP_S.o.). Right, the Δefp_S.o. deletion strain and after complementation with plasmids encoding His₆ versions of EF-P_S.o. (+R32) or the corresponding substitution variants EF-P_S.o.^R32A (+R32A) and EF-P_S.o.^R32K (+R32K), respectively. The presented growth curves are average data from three independent data sets, with statistical error below 10%. (b) β-galactosidase (β-gal) activity assay of S. oneidensis MR-1 Δefp_S.o., ΔearP_S.o. encoding a constitutively produced LacZ-hybrid without (black bars) or with (gray bars) a polyproline motif (3× Pro). β-galactosidase activity is given in percent and is normalized to the wild-type values. The relative activities are average data from three independent data sets, with statistical error below 10%.

EarP is sufficient to activate EF-P

To test whether EarP_S.o. is sufficient for activation of EF-P_S.o., we examined the phenotypes of an E. coli efp deletion (Δefp_E.c.) heterologously producing EF-P_S.o. and EarP_S.o.. As a readout, we used a chromosomal P_cadBA-dependent lacZ reporter in which activation of P_cadBA strictly depends on the pH-responsive transcriptional activator CadC²⁶. Translation of CadC is impaired in cells lacking active EF-P because of the presence of a polyproline motif (Supplementary Fig. 6b)⁷. As expected, the E. coli Δefp_E.c. P_cadBA::lacZ strain was characterized by low-level β-galactosidase activity, and wild-type–like lacZ expression was restored in the presence of a plasmid expressing EF-P_E.c.. Whereas β-galactosidase activity remained low when EF-P_S.o. or EarP_S.o. were expressed alone, simultaneous production of both proteins complemented for the lack of EF-P_E.c.. The diminished growth rate of the Δefp_E.C. mutant¹⁸ was consistently eliminated when EF-P_S.o. and EarP_S.o. were produced together, regardless of the presence of the E. coli EF-P modification enzymes EpmA or EpmB (Supplementary Fig. 6c). Next, we asked whether EpmABC and EarP had adapted to specifically activate their corresponding EF-Ps. Therefore EF-P_E.c. K34 and EF-P_S.o. R32 substitution variants were produced in Δefp_E.c., and P_cadBA activation, growth rate or both were investigated in the resultant strains (Supplementary Fig. 6b,c). Neither EF-P_S.o^R32K with EpmABC nor the EF-P_E.c.^K34R with EarP could reverse the mutant phenotype, thus further corroborating our hypothesis for the coevolution of EF-P with its associated modification system.

EarP modifies EF-P at Arg32

Having demonstrated that EarP_S.o. is necessary and sufficient for specific activation of EF-P_S.o., we addressed whether EarP_S.o. post-translationally modifies the conserved R32 of EF-P_S.o.. To that end, we overproduced His₆-tagged EF-P_S.o. in two S. oneidensis strains, Δefp_S.o. and Δefp_S.o./ΔearP_S.o.. Purified EF-P_S.o. from these two strains was proteolytically digested, and the resulting peptides were analyzed by high-resolution LC/MS/MS using an unbiased ‘dependent peptide’ search²⁷. We detected eight high-confidence R32-containing peptides that were 146.058 Da heavier than their unmodified counterparts (Fig. 3 and Supplementary Fig. 7). This mass shift only occurred on R32-containing peptides of EF-P_S.o. and never when EF-P_S.o. was produced in cells lacking EarP (Supplementary Fig. 8). The fragmentation pattern of the modified peptides strongly suggested the modification site to be on R32 (Fig. 3). To further confirm this, we performed a standard variable modification search with a potential arginine mass shift of 146.058 Da, corresponding to a molecular composition of C₆H₁₀O₄ (146.0579 Da). This second analysis identified the modification to exist exclusively on R32 of EF-P, confirming both the molecular composition and the location of the modification (Supplementary Fig. 9). Notably, no peptides with this modification could be found when two arginine substitution EF-P variants were analyzed (His₆-EF-P_S.o.^R32K and His₆-EF-P_S.o.^R32A; Supplementary Fig. 10). This observation explains the efp-null mutant phenotype of strains encoding His₆-EF-P_S.o.^R32K or His₆-EF-P_S.o.^R32A substitution variants (Fig. 2a and Supplementary Fig. 6) as a consequence of the absence of EF-P_S.o. R32 modification by EarP_S.o..

Figure 3. Dependent peptide MS analysis of the S. oneidensis MR-1 EF-P modification.

C-terminally His₆-tagged EF-P_S.o. was analyzed upon homologous overproduction. Modified R32-containing ‘dependent’ peptides (146.058 Da heavier) were identified by a characteristic ΔM mass shift of the precursor and several fragment ions when compared to the unmodified ‘base peptide’. MS/MS spectra of the best-scoring base peptide-dependent peptide pair are shown. Peak colors: red, blue and light blue refer to y, b and a ions (according to Roepstorff-Fohlmann-Biemann nomenclature), respectively; light green, molecular ion; yellow, neutral losses; purple, internal fragments; green, immonium ions, black peaks, unassigned.

EarP is a rhamnosyltransferase

We reasoned that a modification on R32 with a molecular composition of C₆H₁₀O₄ might be the result of N-glycosylation with an activated deoxyhexose sugar (C₆H₁₂O₅ – H₂O = C₆H₁₀O₄). E. coli synthesizes two nucleotide diphosphate deoxyhexoses: GDP-l-fucose and dTDP-l-rhamnose (Fig. 4a)²⁸. The biosynthesis genes encoding the latter (represented by rmlD) are strictly conserved in bacteria encoding EarP but are less frequently found in bacteria with EpmABC (Supplementary Fig. 5). To test whether dTDP-l-rhamnose or GDP-l-fucose might act as a substrate for EarP_S.o., we interrupted the corresponding synthesis pathways. GDP-l-fucose and dTDP-l-rhamnose are synthesized from fructose-6-phosphate and glucose-l-phosphate, respectively, with each biosynthetic pathway encompassing four specific enzymes (Fig. 4a). To prevent formation of GDP-l-fucose, we deleted fcl, which arrested synthesis at the GDP-4-keto-6-deoxy-d-mannose step, whereas dTDP-l-rhamnose formation was blocked at the intermediate dTDP-4-keto-6-deoxy-l-mannose step by deletion of rmlD (Fig. 4a). The deletions were generated in an E. coli P_cadBA::lacZ strain lacking efp. The β-galactosidase activities of both resultant strains deficient either in GDP-l-fucose (Δefp_E.c./Δfcl_E.c.) or in dTDP-l-rhamnose (Δefp_E.c./ΔrmlD_E.c.) were comparable to wild-type activity when a copy of efp_E.c. was provided in trans (Fig. 4b). Similarly, the Δefp_E.c./Δfcl_E.c. strain could be complemented with EF-P_S.o./EarP_S.o., indicating that GDP-l-fucose is not a substrate for EarP. In stark contrast, the coproduction of EF-P_S.o./EarP_S.o. in the context of Δefp_E.c./ΔrmlD_E.c. cells phenocopied the Δefp_E.c. strain, suggesting that dTDP-l-rhamnose is the substrate needed for EarP_S.o. to activate EF-P_S.o. To further corroborate this result, we deleted rmlC (locus tag SO_3160) in S. oneidensis and analyzed translation of a polyproline-containing LacZ reporter hybrid in the resultant dTDP-l-rhamnose–deficient strain ΔrmlC_S.o. (Fig. 4c)^3,7. As observed for Δefp_S.o. or ΔearP_S.o., β-galactosidase activity was low in the ΔrmlC_S.o. strain harboring the p3LC-TL30-3P reporter. Consistent with our previous results, in vivo activation of EF-P_S.o. strictly depends on the biosynthesis of dTDP-l-rhamnose, leading us to conclude that dTDP-l-rhamnose acts as the substrate for EF-P_S.o. modification and that it was recruited for this role upon the development of the EarP–EF-P phylogenetic relationship (Supplementary Figs. 4,5 and Supplementary Data Set 1).

Figure 4. In vivo analysis of S. oneidensis MR-1 EF-P functionality depending on NDP-deoxyhexose biosyntheses.

(a) Biosynthesis pathways for dTDP-l-rhamnose, GDP-d-rhamnose and GDP-l-fucose. Arrows depict sugar conversion steps. Specific conversion steps are associated with the corresponding biosynthesis gene. Paralogous genes are separated by a comma, and alternative names are given in parentheses. (b) β-galactosidase (β-gal) activity of E. coli P_cadBA-lacZ reporter wild-type (WT) and efp deletion strain (Δefp_E.c.) as well as the Δefp_E.c./Δfcl_E.c. (GDP-l-fucose deficient) and Δefp_E.c./ΔrmlD_E.c. (dTDP-l-rhamnose deficient) double deletion mutants and after complementation either with EF-P_E.c. or EF-P_S.o. in combination with EarP_S.o.. Cells were incubated under cadBA-inducing conditions. Data represent mean values from three independent replicates ± s.d. (c) β-galactosidase activity assay of S. oneidensis MR-1 ΔrmlC_S.o. encoding a constitutively produced LacZ hybrid without (black bars) or with (gray bars) a polyproline motif (3× Pro). β-galactosidase activity is given in percent and is normalized to the wild-type values. The relative activities are average data from three independent data sets, with statistical error below 10%.

To directly demonstrate that EarP_S.o. can glycosylate EF-P_S.o. using dTDP-l-rhamnose as a substrate, we performed in vitro glycosylation reactions with purified components. LC/MS/MS analysis, performed as described above, revealed the presence of R32 rhamnosylation of wild-type EF-P_S.o. if and only if all three components were provided (Supplementary Fig. 11). Collectively, our data demonstrate unambiguously both in vivo and in vitro that EarP is an EF-P arginine rhamnosyltransferase essential for post-translational activation.

EF-P stimulates peptide bond formation indirectly

Rhamnosyl-arginine differs substantially from (R)-β-lysinylhydroxylysine and hypusine of EF-P_E.c. and a/eIF5A, respectively, raising the question of how this unusual extension protrudes into the peptidyl-transferase center of the ribosome (Fig. 5a–d). To investigate this, we generated molecular models for the different modifications of EF-P orthologs based on the crystal structure of unmodified Thermus thermophilus EF-P bound to the 70S ribosome¹². These models suggest that the (R)-β-lysinylation found on EF-P_E.c. could reach within 2 Å of the proline attached to the P-site tRNA (Fig. 5b), whereas the hypusine and rhamnose-arginine modifications are shorter and cannot reach the P-site proline (Fig. 5c,d). Therefore, EF-P bearing either hypusine or rhamnosearginine modifications is not likely to stimulate peptide-bond formation by directly influencing the conformation of the polypeptide chain but rather does so indirectly by interacting with and stabilizing the CCA-end of the P-site peptidyl-Pro-tRNA.

Figure 5. EF-P rhamnosylation, mode of action and impact on pathogenicity.

(a) Arginine-rhamnosylation by EarP using dTDP-l-rhamnose as substrate. (b–d) Models of different modified EF-P proteins bound to the ribosome. The CCA-end of P-site bound Pro-tRNA (blue) is shown for reference. Models based on T. thermophilus EF-P-70S structure¹². (b) K34 of E. coli EF-P post-translationally modified with (R)-β-lysine. (c) EF-P bearing the K34 hypusine modification. (d) R32 of EF-P modified by l-rhamnose. (e) Effects of Δefp_P.a. and ΔearP_P.a. on P. aeruginosa pathogenicity. Cytotoxicity of P. aeruginosa strains was assessed by infecting A549-Gluc cells, which secrete Gaussia luciferase, as a measure of cell integrity. Data represent mean values from three independent replicates ± s.d.

EarP and EF-P are essential for P. aeruginosa pathogenicity

Distinct bacterial strategies to functionalize EF-P may provide a basis for development of customized antibiotics. Deleting EF-P or its modifying enzymes has been shown to reduce bacterial fitness^16,25 and lead to a loss of pathogenicity in Salmonella enterica and A. tumefaciens. To test whether rhamnosylated EF-P is also required to develop pathogenicity in the P. aeruginosa strain PAO1, we investigated transposon mutants of efp_P.a. (locus tag PA2851) and earP_P.a. (locus tag PA2852) in an infection assay using the human cell line A549-Gluc (Fig. 5e). Whereas wild-type P. aeruginosa decreased the number of living cells by about 80%, infection with Δefp_P.a. or ΔearP_P.a. mutants had no effect on cell viability. Pathogenicity of P. aeruginosa PAO1 is dependent on a large number of cell-associated and extracellular virulence factors, such as rhamnolipids and pyocyanin, that are important for colonization and invasion during infection²⁹. Bioinformatic analysis on the P. aeruginosa proteome revealed that the synthesis of those virulence factors involves polyproline-containing proteins, suggesting a dependence on EF-P for their translation (Supplementary Table 4). Consistently, efp_P.a. or earP_P.a. disruption mutants showed a substantial decrease in the production of rhamnolipids and pyocyanin, and production was restored by introducing EarP_S.o. and EF-P_S.o. but not the substitution mutants EF-P_S.o.^R32A and EF-P_S.o. ^R32K (Supplementary Fig. 12). Therefore, both EF-P_P.a. and the corresponding rhamnosyltransferase EarP_P.a. contribute to pathogenicity in P. aeruginosa.

DISCUSSION

Protein glycosylation is a commonly used strategy to alter structural and functional properties of a protein. However, until recently, N-linked glycosylation was almost exclusively associated with asparagine. The only known additions of a sugar to arginine were restricted to two reported examples on eukaryotic proteins. Arginine glycosylation was discovered first in search of a protein primer for starch synthesis in 1995 (ref. 30). Here, the authors identified sweet corn amylogenin to be self-β-glucosylated. Second, in 2013, two independent research groups showed that NleB, an enteropathogenic E. coli type III secretion system effector, antagonizes death receptor signaling by modifying conserved arginines in human death receptor domains with N-acetylglucosamine (GlcNAc)^31,32. With S. oneidensis and P. aeruginosa EF-P, we now report what is to our knowledge the first arginine-glycosylated bacterial protein, thus demonstrating that this type of post-translational modification is not restricted to eukaryotic proteins but is common to other domains of life.

ONLINE METHODS

Bioinformatics software

Hidden Markov model (HMM) analyses were carried out using the HMMER3 software package³³. Multiple sequence alignments were constructed using the l-ins-i algorithm of the MAFFT version 6.864b software package³⁴. BLASTP searches were performed using the BLAST+ software package version 2.2.26 (ref. 35). All phylogenetic trees were constructed using FastTree with default settings³⁶. Sequences logos were created using the Weblogo server³⁷.

Genome set and domain architecture

1,611 completely sequenced prokaryote genomes from a previously defined set available from 4 April 2012 were collected³⁸. We used a set of 1,004 genomes with reduced redundancy based on the 16S comparison for all sequence analyses. The Pfam26 HMM library³⁹ was used to define domain architecture of all sequences with default gathering thresholds. In the event of domain overlaps, the highest-scoring domain model was chosen for the final architecture.

EarP, EF-P, RmlD and EpmC were identified by collecting sequences that contain the DUF2331, EF-P, RmlD_sub_bind and DUF462 domains, respectively. EpmA consists of a class II tRNA synthetase domain (tRNA-synt_2), which is found in a variety of proteins. All 2,684 sequences containing the tRNA-synt_2 domain were collected and aligned. The core region corresponding to the tRNA-synt_2 domain was extracted from the multiple sequence alignment and used to build a phylogenetic tree. Domain architecture and gene neighborhood analyses identified a conserved clade of 239 sequences in the tree that were determined to be EpmA homologs because of their genomic context association with EF-P homologs (Supplementary Fig. 1). EpmB is composed of a Radical SAM domain (Radical_SAM), which is found in over 15,000 proteins in our sequence set, making a phylogenetics-based approach to EpmB identification challenging. EpmB is often encoded near efp or epmA. We collected all of the sequences with Radical_SAM domains that were encoded within a distance of four genes to efp or epmA homologs (181 sequences) and aligned them. A tree constructed from the alignment revealed two distinct clades, one with many short branches presumed to be true EpmB sequences and another composed of many long branching sequences, which suggest divergence. Furthermore, some sequences in the divergent clade are from genomes represented in the EpmB-associated clade, supporting that they are not true EpmB orthologs (Supplementary Fig. 2). These 36 sequences of the divergent clade were removed from the set, and the remaining sequences were used as queries in BLASTP searches against our representative sequence set. All sequences with an e-value of 0.0001 or less were collected (515 sequences) and aligned. A phylogenetic tree was built from the core region of the multiple sequence alignment corresponding to the Radical_SAM domain. Conserved clades that were associated with EF-P or EpmA on the basis of genome context and phylogenetic distribution were identified (Supplementary Fig. 3). The 236 sequences that are members of these clades were defined as EpmB homologs, and this is consistent with the 239 EpmA homologs we identified that are presumably part of the same pathway. All EarP, EF-P, RmlD, EpmC, EpmA and EpmB homologs in the representative genome set can be found in Supplementary Data Set 1. We further identified all of the EarP homologs in the full 1,611 genome set on the basis of the presence of the DUF2331 domain (Supplementary Data Set 2). We identified all of the EarP-associated EF-Ps in this set using a HMM built from an alignment of the EarP-associated EF-Ps identified in the EF-P phylogenetic analyses. All sequences with a score greater than or equal to that of the lowest-scoring member of the representative set of EarP-associated EF-Ps were identified as homologs (Supplementary Data Set 2).

Phylogenetic analysis of EF-P homologs

A phylogenetic tree was built from the core region of a multiple sequence alignment of EF-P homologs. Conserved clades associated with EarP, EpmA or EpmB on the basis of genome context and phylogenetic distribution data were identified (Supplementary Fig. 4). Sequences from these clades were collected, with the exception of those from an EpmA/EpmB-associated subfamily that includes members of the EF-P-like family (the YeiP subfamily), which lack the conserved lysine. The collected sequences were aligned, and the core region of the alignment was used to construct a phylogenetic tree (Supplementary Fig. 5).

Oligonucleotides, plasmids and bacterial strain construction

Primers, plasmids and strains used in this study are listed in Supplementary Tables 1–3. Of note, transposon mutant P. aeruginosa PAO1 ID 30853 from the Washington Genome Center⁴⁰ with a transposon insertion in open reading frame (ORF) PA4684 was used as a wild-type control⁴¹. Strains BW-Δefp/epmA::npt, BW-Δefp/epmB::cat, MG-CR-efp-fcl and MG-CR-efp-rmlD were constructed by using pRED/ET recombination technology together with rpsL counter-selection⁴² in accordance to the technical protocol of the Quick and Easy E. coli Gene Deletion Kit of Gene Bridges (http://www.genebridges.com/). Strains Δefp_S.o., ΔearP_S.o., Δefp_S.o./ΔearP_S.o. and ΔrmlC_S.o. were constructed as essentially described in ref. 43, leaving terminal sections of the target gene.

Molecular biology methods

Enzymes and kits were used according to the manufacturer's directions. Genomic DNA was purified according to standard protocols. DNA fragments were purified from agarose gels using a high-yield PCR cleanup and gel extraction kit (Sued-Laborbedarf). Restriction endonucleases were purchased from New England Biolabs. Sequence amplifications by PCR were performed by using the Phusion high-fidelity DNA polymerase from Finnzymes or the Taq DNA polymerase from New England Biolabs, respectively. All efp mutants were constructed by one- or two-step PCR using mismatched primer pairs⁴⁴.

Growth conditions

E. coli, P. aeruginosa PAO1 and S. oneidensis MR-1 were routinely grown at 37 °C (for E. coli and P. aeruginosa) and 30 °C (S. oneidensis), unless indicated otherwise. According to the NaCl modification of Miller, lysogeny broth (LB)⁴⁵ was used as complex medium. When indicated, LB was buffered with 100 mM sodium-phosphate to pH 5.8. Microaerobic conditions were achieved by growing cells in closed Eppendorf cups with minimal agitation. Antibiotics were used when necessary with the following concentrations: 100 μg/ml ampicillin sodium salt, 50 μg/ml kanamycin sulfate, 34 μg/ml chloramphenicol, 50 μg/ml streptomycin sulfate or 15 μg/ml tetracycline hydrochloride. For blue-white selection, LB agar plates were additionally supplemented with 80 μM 5-bromo-4-chloro-3-indolyl β-d-galactopyranoside (X-Gal; Sigma Aldrich) and 1 mM isopropyl β-d-1-thiogalactopyranoside (IPTG; Sigma Aldrich).

β-Galactosidase activity assay

Cells expressing lacZ under control of the cadC or cadBA promoters were grown in buffered LB medium to mid-exponential growth phase or overnight and harvested by centrifugation. β-Galactosidase activities were determined for at least three independent experiments and are given in Miller units (MU). The significance of the results was determined by applying the two-sided Student's t-test, and results were considered significantly different if P < 0.05.

Overproduction and purification of recombinant proteins

C-terminal His₆-tagged EF-P_S.o. as well as the corresponding R32A and R32K substitution variants were overproduced either in S. oneidensis MR-1 wild-type, Δefp_S.o. and Δefp_S.o./ΔearP_S.o. mutant strains or in E. coli LMG194 and grown in LB overnight at 16 °C after the addition of 0.2% (w/v) l-arabinose to exponentially grow cells. Similarly, C-terminal His₆-tagged EarP_S.o. was produced in E. coli LMG194. Cells were lysed and purified using Ni-NTA (Qiagen) and 250 mM imidazole. For further MS analysis and in vitro rhamnosylation, proteins were dialyzed against reaction buffer (50 mM HEPES, pH 7.8, 100 mM NaCl, 50 mM KCl, 10 mM MgCl₂, 2.5 mM β-mercaptoethanol, 2.0% (v/v) glycerol).

Protein digestion and sample preparation for MS

Purified EF-P was predigested with LysC for 3 h in 8 M urea, 50 mM Tris-HCl, pH 7.5, 1 mM DTT and 5 mM chloroacetamide (CAA). Then, samples were diluted 1:4 with 50 mM Tris HCl, pH 7.5, to decrease the urea concentration to 2 M and digested overnight with trypsin. Peptide mixtures were purified on C18 StageTips⁴⁶.

LC/MS/MS analysis

Peptides were eluted from the C18 StageTips according to the standard protocol. They were analyzed by reversed-phase LC on an EASY-nLC 1000 system (Thermo Fisher Scientific) directly coupled to a quadrupole Orbitrap mass spectrometer (Q Exactive, Thermo Fisher Scientific). HPLC columns with a length of 50 cm and an inner diameter of 75 μm were packed in-house with ReproSil-Pur 120 C18-AQ 1.9-μm particles (Dr. Maisch GmbH). Peptide mixtures were separated using gradients of either 60 min or 140 min (total run time plus washout) and a two-buffer system: buffer A++ (0.1% formic acid) and buffer B++ (80% acetonitrile in 0.1% formic acid). The flow rate was set to 250 nl/min, and the column was heated to 50 °C using a column oven (Sonation GmbH). Peptides eluting from the column were directly sprayed into the mass spectrometer; spray voltage was set to 2.3–2.4 kV, and the capillary temperature was set to 250 °C. The mass spectrometer was operated in a data-dependent mode with switching between a survey scan and fragmentation scans of the top five most abundant peaks. In the 60-min method, full scans were acquired at a resolution of 140,000 with an AGC target of three E06 ions and a maximum injection time of 20 ms. Precursors were selected with an isolation window of 3 Th, and MS2 scans were acquired at a resolution of 17,500 with an AGC target of one E05 ion and a maximum injection time of 120 ms. In the 140-min method, full scans were acquired at a resolution of 70,000 with an AGC target of three E06 ions and a maximum injection time of 20 ms. Precursors were selected with an isolation window of 2 Th, and MS2 scans were acquired at a resolution of 35,000 with an AGC target of one E05 ion and a maximum injection time of 120 ms. In both cases, peptides were fragmented by higher-energy collisional dissociation (HCD) with a normalized collision energy of 25. To minimize resequencing of peptides, dynamic exclusion was enabled within a time window of 20 s.

MS data analysis

MS raw files were processed using MaxQuant²⁷ version 1.5.0.0. MS/MS spectra were searched using the Andromeda search engine⁴⁷ against FASTA files obtained from Uniprot adapted to the corresponding sample. For EF-P samples from homologous production in S. oneidensis, raw data were searched against the S. oneidensis reference proteome downloaded from Uniprot on 20 January 2014 and a FASTA file containing the sequence of His₆-tagged EF-P_S.o.. Depending on the experiment, we added additional FASTA files containing the sequence of His₆-tagged EF-P_S.o.^R32A or EF-P_S.o.^R32K. Cysteine carbamidomethylation was set as a fixed modification; N-terminal acetylation and methionine oxidation were set as variable modifications. Trypsin was chosen as the specific enzyme, with a maximum of two missed cleavages allowed. Peptide and protein identifications were filtered at a 1% false discovery rate (FDR). The initial mass tolerance was set to 4.5 p.p.m. for the precursor masses and to 20 p.p.m. for the fragment masses. All of the other parameters were left at standard settings. For dependent peptide analysis (essentially as described in ref. 47), the corresponding feature was enabled. For the variable modification searches, l-rhamnose –H₂O (C₆H₁₀O₄ = 146.058 Da) was first defined as a variable modification with specificity for arginine in the Andromeda con figuration and was subsequently added to the other variable modifications in the MaxQuant search.

Bioinformatic analysis of the MaxQuant processed data was performed using the Perseus software (version 1.4.2.35, available in the MaxQuant environment). In brief, for dependent peptide analysis, the ‘all.peptides.txt’ table was loaded and filtered for DP decoy ≠ ‘+’, ‘DP Protein’ = EF-P, ‘DP Base sequence’ containing ‘SGR’, ‘DP Mass Difference’ > 0 and ‘DP Score’ > 80. Remaining hits were further validated in a manual fashion. Spectra were visualized using the Viewer program (version 1.5.0.0, integrated into MaxQuant) and annotated using the Expert System⁴⁸. For rhamnosylation analysis, the ‘Rhamsites.txt’ table was loaded, and the site table was expanded, logarithmized and then filtered for ‘Protein’ = EF-P, ‘Localization Probability’ = 1 and ‘Score’ > 80.

Enzymatic total synthesis of TDP-l-rhamnose

The synthesis was carried out in two steps. First, TDP-4-keto-6-deoxy-d-glucose was prepared from glucose-1-phosphate using two purified E. coli enzymes (RmlA and RmlB) and TTP. TTP was generated in situ from TMP with a mix of TMP kinase and acetate kinase. TDP-4-keto-6-deoxy-d-glucose is a key intermediate of many 6-deoxyhexoses and can be stored at –80 °C. E. coli BL21(DE3) (EMD 4 Biosciences) was used for the conversion of TDP-4-keto-6-deoxy-d-glucose to TDP-l-rhamnose. This E. coli strain contains naturally TDP-4-ketorhamnose 3,5-epimerase (RmlC) and TDP-4-keto-rhamnose reductase (RmlD). Crude cell lysates of E. coli BL21 (DE3) were added to a solution of TDP-4-keto-6-deoxy-d-glucose in phosphate buffer (50 mM, pH 7.5), MgCl₂ (4 mM) and NADH (5 mM). The reaction continued for 3 h at 37 °C. The production of TDP-l-rhamnose was monitored by HPLC^49,50.

The resulting solution containing TDP-l-rhamnose was desalted by size-exclusion chromatography (BioGel P2 column) and further purified by HPLC (Waters 600 system consisting of a controller, a Waters 996 photodiode array detector and a Delta 600 pump; a Dionex CarboPac PA1, 4 × 250 mm column was used for 60-min runs at a flow rate of 1.0 ml/min, and UV monitoring absorbance was set at 254 nm). The gradient used was as follows: solvent A, water; solvent B, 0.5 mM ammonium acetate solution. Solvent B was increased from 5% to 20% (0 min to 15 min), from 20% to 60% (15 min to 35 min), and then from 60% to 100% (35 min to 45 min); it was kept at 100% for 5 min before it was decreased back to 5% within 2 min, and it was kept at 5% for the last 8 min^49,51. Under these conditions, TDP-l-rhamnose elutes at 34.04 min.

In vitro glycosylation

A total of 10 μM of His₆-EF-P_S.o. or substitution variants and 10 μM His₆-EarP_S.o. were incubated in reaction buffer (50 mM HEPES, pH 7.8, 100 mM NaCl, 50 mM KCl, 10 mM MgCl₂, 2.5 mM β-mercaptoethanol, 2.0% (v/v) glycerol) with 100 μM of dTDP-l-rhamnose for 60 min at 30 °C.

Cytotoxicity assay

The cytotoxicity of the P. aeruginosa strains was assayed as described earlier⁵². Briefly, eukaryotic A549-Gluc cells were cultured in completed Dulbecco's modified Eagle's medium at 37 °C with 5% CO₂. A549-Gluc cells were generated from A549 by lentiviral gene transfer as described previously^53,54. Cytotoxicity of the P. aeruginosa strains was assessed by infecting A549-Gluc cells, which secrete Gaussia luciferase, as a measure of cell integrity. A549-Gluc cells were seeded in 96-well plates at a density of 2.5–5 × 10⁴ cells per well and grown until ~90% confluence. After washing, cells were inoculated with 6-h-old P. aeruginosa LB cultures adjusted to a multiplicity of infection (MOI) of 200 and centrifuged to increase cell-cell contact. 100 μl cell culture super natants were collected after 3 h of incubation, and Gaussia luciferase activity was measured for 0.1 s using an LB 960 Centro XS3 plate luminometer (Berthold Technologies) after the addition of 60 μl of 10 μM coelenterazine (PJK GmbH). Luciferase activities were determined for at least three independent experiments. The significance of the results was determined by applying the two-sided Student's t-test, and results were considered significantly different if P < 0.05.

Assay for pyocyanin production

Pyocyanin was extracted from P. aeruginosa supernatant and measured according to ref. 55. 5 ml of 24-h-old cultures were extracted with 1 volume of chloroform and then re-extracted into 0.2 N HCl to give a pink solution. The aqueous layer was transferred to a fresh tube, and absorbance was measured at 520 nm. Pyocyanin produced per milliliter of culture supernatant was calculated as described elsewhere⁵⁵.

Quantification of rhamnolipid production

Overnight cultures of P. aeruginosa were freshly diluted to an OD_600nm of 0.05 and incubated in LB at 37 °C for 48 h. The colorimetric analysis of the orcinol reaction was adopted from the method described in ref. 56. Briefly, 300 μl of culture super-natant were extracted twice with diethylether. After evaporation of the pooled fraction, the remaining fraction was dissolved in distilled water and incubated with 100 μl 1.6% (w/v) orcinol and 800 μl 60% sulfuric acid. After heating to 80 °C and shaking at 175 r.p.m. for 30 min, the adsorption at 421 nm was determined. In parallel, l-rhamnose at defined concentrations was assayed as described above and used as a standard for determining the l-rhamnose in the culture samples. Rhamnolipid concentrations were then calculated with the assumption that 1 μg of l-rhamnose corresponds to 2.5 μg of rhamnolipid⁵⁷. Rhamnolipids were quantified from at least three independent experiments. The significance of the results was determined by applying the two-sided Student's t-test, and results were considered significantly different if P < 0.05.

Molecular modeling

The molecular model for modified EF-Ps on the ribosome was generated using the crystal structure of unmodified T. thermophilus EF-P bound to a T. thermophilus 70S ribosome programmed with tRNA^fMet at the P-site¹². The proline residue was modeled onto the CCA-end of the P-site tRNA by aligning the structure of the 50S subunit with an aminoacylated tRNA substrate in the P-site (PDB1VQN) and by mutagenesis of the amino acid to proline using Coot⁵⁸. The models for EF-P bearing lysinylation or hypusine modifications were generated by mutagenesis of Arg32 to lysine of ribosome-bound T. thermophilus EF-P and then addition of the required modification moieties to the ε-amino group of the lysine, whereas for EF-P bearing the rhamnosylation modification, an l-rhamnose moiety was added to the η²-amino group of Arg32 of ribosome-bound T. thermophilus EF-P. All of the models were generated and refined in Coot⁵⁸, and images were produced using the PyMOL Molecular Graphics System (Version 1.5.0.4 Schrödinger, LLC).