EF-P translation stall rescue

Species-neutral bacterial module for elongation factor P-dependent rescue of cytosolic ribosomes that stall while translating consecutive proline residues. EF-P is universal in bacteria and is the ortholog of archaeal aIF5A and eukaryotic eIF5A, whose hypusine-activated stall rescue is the same biology in another domain of life and is modelled separately. The module has three roles: lineage-specific activation of EF-P by post-translational modification of its loop-1 tip residue, engagement of the stalled ribosome, and stimulation of peptide-bond formation so elongation resumes. The activating chemistry differs by lineage and is modelled as a variant set over three characterized routes: EpmA/EpmB/EpmC (R)-beta-lysylation and hydroxylation of Lys34 in Escherichia coli and Salmonella, EarP rhamnosylation of Arg32 in Pseudomonas and Shewanella, and 5-aminopentanolylation of Lys32 in Bacillus subtilis.

Boundary and generalization decision (2026-11-02). The reusable module is the bacteria-wide EF-P stall-rescue mechanism, not the P. putida instance: the earlier draft was written around the local UniPathway UPA00345 bucket and named KT2440 in its scope, which made a universal factor look pseudomonad- specific. KT2440 efp is now one of three reviewed exemplars (with E. coli P0A6N4 and B. subtilis P49778), all in PANTHER subfamily PTHR30053:SF12, and the module label, description and context are species-neutral. The activating modification was previously declared adjacent context. That was the wrong call for a reusable module: activation is what makes EF-P competent for the rescue step in most lineages, and the three characterized chemistries are alternative implementations of one role, which is exactly what `variant_sets` are for. They are modelled as a ROUTE variant set (the activities differ: an aminomutase plus a lysyl ligase plus a hydroxylase, a rhamnosyltransferase, or a carbonyl reductase), with EXACTLY_ONE selection because no organism is known to run two of them. The enclosing part is `optional` because B. subtilis EF-P retains function when unmodified and many bacteria have no identified EF-P modification system at all, so an organism that satisfies the EF-P step without any activation route is a real case, not a gap. What remains outside the module: dTDP-L-rhamnose supply (rml/rfb) for the EarP route, the free-lysine supply for EpmB, the upstream genes that build the B. subtilis 5-aminopentanone donor (ynbB, gsaB), and the downstream phenotypes that EF-P loss produces (swarming motility, acid resistance, virulence). These are substrate supply, separate pathways, or consequences of translating individual polyproline proteins, not parts of the rescue mechanism. EfpL/YeiP is deliberately outside the module core. It is a real paralog that resolves an overlapping but not identical set of arrest motifs, and its own loop-tip arginine is unmodified, so it is neither a second copy of the EF-P role nor a user of the activation routes modelled here; organisms that encode it should have it modelled explicitly. EpmB asserts no function and no process. It supplies free (R)-beta-lysine and never touches EF-P, so by the participation test in CLAUDE.md it is not involved in the modification process; its own gene review marks GO:0043687 over-annotated for that reason, and asserting the term here would have contradicted the review in machine-readable form. The donor-supply role is carried by role_description until GO has a term for the reaction. Compartment granularity is deliberately mixed. The module context and the EF-P, EpmA and EarP annotons use GO:0005829 cytosol, which their gene reviews support (E. coli EF-P and EpmA each have a cytosol row with direct evidence). The EpmC and YmfI annotons use the parent GO:0005737 cytoplasm, because those two reviews add cytoplasm as a curator inference (IC) with no fractionation data behind it; claiming cytosol for them would assert more than the evidence carries. GO:0072344 is the module's central process term and is the term GO and the EF-P/eIF5A literature use, but its definition speaks of freeing the ribosome from the stalled translation complex, whereas EF-P relieves the stall so that the same ribosome resumes elongation. The term is kept because it is the established one and no better term exists; the BACSU efp review raises the definition mismatch as a question for the ontology, and this module inherits that caveat rather than inventing a replacement. Member review status. Seven of the eight member reviews are status COMPLETE. PSEPK efp is left at DRAFT deliberately: its content meets the bar (all seven annotations reviewed, core function present, validation passes), but its only outstanding item is that no annotation cites its deep-research file, and that file (efp-deep-research-asta.md) is a retrieval artifact whose papers are about other subjects entirely, with nothing about EF-P to quote. Citing it to clear the warning would be padding, so the warning stands and the status is left for whoever re-runs provider research on that gene. Grounding choices. The EF-P roles are grounded on PTHR30053:SF12, verified in interpro/panther/panther.obo and in the PANTHER member index, with all three exemplars classified at exactly that subfamily. The activation enzymes are grounded as concrete UniProt gene products instead of by family, because each of their PANTHER families (PTHR42918 for EpmA, PTHR30538 for EpmB, PTHR42879 for YmfI) is dominated by a different activity and would make a false family-level claim; all three families nevertheless have full reviews recording that judgment, and EpmC has no PANTHER family at all. EpmB's annoton carries no GO molecular function because GO has no term for the (R)-beta-lysine-forming reaction (RHEA:44148); the epmB gene review proposes one, and the id should be filled in here once GO creates it.