Bacteria Beat Programmable Antibiotics at the Peptide Door

Bacteria exposed to programmable antibiotics called asobiotics evolve resistance mainly by shutting the peptide door that carries the drug inside four Gram-negative pathogens. A TWINCORE-led team reported the pattern on 12 August 2026 in Nature Communications, after laboratory evolution on Escherichia coli, Klebsiella pneumoniae, Salmonella enterica, and Pseudomonas aeruginosa.

Adam Mulkern, a former postdoctoral researcher in TWINCORE’s Systems Biology of Microbial Communities group and first author, said the bugs “either prevent the antibiotic from entering the cell or alter the cellular response once the molecule reaches its target.” Which path they take depends on the peptide used to haul a short antisense oligomer across the membrane.

Sixteen Passages Forced Resistance Into View

Asobiotics are antisense oligomers written to sit on the start of an essential bacterial messenger RNA and block the ribosome from making the protein. The chemistry is usually a peptide nucleic acid, or PNA, or a morpholino oligomer, each paired with a cell-penetrating peptide because the oligomer alone does not cross a Gram-negative envelope. Jörg Vogel’s group at the University of Würzburg has framed that pairing as a route to RNA-directed antisense oligonucleotides for microbiota editing as well as for drug-resistant infections, with the sequence chosen by simple base-pairing rules.

The new paper, with Lisa Popella and Thu-Hien Vu among the co-authors, tests that promise the way microbiologists test any new drug: they try to break it. Marco Galardini, the TWINCORE group leader and senior author, and colleagues ran a “drug selection ramp,” 16 daily 1:100 passages that started at half the minimum inhibitory concentration and doubled the dose every four passages, finishing at four times the starting MIC. Ten biological replicates were run for each strain-and-drug pair.

They used two PNA sequences, one aimed at acpP, a fatty-acid gene, and one at rpsH, a ribosomal-protein gene, and three peptides: (KFF)3K in its usual L-form, a D-form of the same peptide, and (RXR)4XB. KFF-acpP went onto every strain except P. aeruginosa, which has no sbmA homolog and does not take up KFF well, so that species got RXR-acpP instead. E. coli MG1655 saw all four constructs. Scrambled PNAs on the same peptides left growth untouched and MICs above 80 µM, so the killing was not a generic peptide effect.

The KFF Peptide Relies on a Breakable Door

KFF is the old workhorse. Earlier work showed that after a periplasmic protease chews the peptide, the inner-membrane protein SbmA hauls the leftover PNA into the cytoplasm. In this ramp, that single door dominated. For every species treated with KFF-acpP, sbmA was consistently mutated across species in at least six of ten replicates. Nine of ten MG1655 lines adapted to KFF-rpsH also picked up sbmA variants. The co-transcribed outer-membrane gene yaiW mutated in seven KFF-acpP lines and four KFF-rpsH lines of MG1655.

Salmonella paid the steepest price on KFF-acpP, climbing from an average ancestral MIC of 1.88 µM by a 17.6-fold jump, the largest mean increase in the study. MG1655 on the same construct went from 5 µM to a 6-fold higher MIC. The same strain on KFF-rpsH rose only 2.5-fold, so the peptide set the path and the oligo still shaped how far resistance ran. An E. coli knockout that already lacked sbmA started at 20 µM on KFF-acpP, four times the MG1655 value, and still found another 2-fold after the ramp.

Adapted populations carried 34.9 genetic variants on average, against 12 in the scrambled controls. P. aeruginosa threw off 199.5 variants per replicate, the noisiest genome in the set. Salmonella often mutated rpoS, in eight replicates, and rfbF, in five, a gene that builds the outer-membrane O-antigen and has already been tied to weaker PNA activity. Klebsiella mutated mdtC, part of a multidrug efflux pump, in four replicates. Those extras sat on top of the KFF uptake defect rather than replacing it.

Eight of Ten Lines Stayed Sensitive to RXR

Switch the peptide and the genetics flip. D-KFF, which resists that periplasmic digest, produced no sbmA mutations at all in E. coli. Mean MIC rose only 1.3-fold, the smallest shift recorded, and a few lines picked up spoT variants that may link the fatty-acid target to a starvation response. RXR, which earlier studies say crosses membranes in a way that depends on membrane potential rather than SbmA, looked even quieter on MG1655: mean MIC rose 1.4-fold, and 8 of 10 adapted populations showed no increase against the ancestral isolate.

Where RXR resistance did appear, it was modest and sat in the PNA binding site on acpP. The authors note those mutations can be bypassed by changing the oligo sequence, which is the move the whole class is built to make. That is the opposite of KFF, where bacteria break the transporter and never let the sequence vote. The paper’s own opening analogy, updating an asobiotic the way an mRNA vaccine is updated, only describes the RXR outcome.

P. aeruginosa, tested only on RXR-acpP because KFF does not deliver well without SbmA, still grew through the ramp. The authors treat many of the high-frequency variants on D-KFF and RXR as possible passenger mutations, because those constructs barely moved the MIC. The practical split is crude and useful: L-form KFF teaches the cell to close a known door; RXR mostly does not.

How a Single prfB Mutation Raises Resistance

Uptake is not the whole KFF story. Three MG1655 KFF-acpP replicates jumped more than 8-fold even though all ten lines had similar sbmA damage. Sample M02 hit a 16-fold MIC increase and carried a private change at codon 246 of prfB, T246S. That gene encodes peptide chain release factor RF2, which helps recover stalled ribosomes. Older biochemical work had already shown that changes at position 246 can raise termination efficiency. A related hit, prfA C127F, turned up in the uropathogenic E. coli strain 536 on the same construct.

The team rebuilt the standout alleles in a clean MG1655 background. A double deletion of sbmA and yaiW reproduced the uptake-class resistance. The reconstructed prfB T246S mutant reached a MIC of 20 µM, a 4-fold rise over the ancestor, matching seven of the ten KFF-acpP evolved lines and doing so with no uptake mutation on board. Asobiotics are taught as steric blocks that stop the ribosome from binding in the first place. A terminator protein still moved the MIC, which is why the authors flag translation-recovery genes as a second, uptake-independent route.

VALIDATED AND RECURRING RESISTANCE ROUTES

• SbmA uptake: Loss-of-function hits in the inner-membrane transporter, often with yaiW, blocked KFF-linked PNA in E. coli, Salmonella, and Klebsiella.
• Release factor: The prfB T246S swap alone took the MIC to 20 µM, and a prfA change appeared in the UPEC strain.
• Envelope and efflux: Salmonella rfbF and Klebsiella mdtC mutations sat in the same ramps, pointing at the outer membrane and a pump complex rather than the oligo sequence.
• Binding-site SNPs: Rare on RXR-acpP, and the paper says a new PNA sequence can walk around them.

ykfM, an uncharacterized transmembrane gene tied to chemical and physical stress, mutated in every MG1655 and ΔsbmA replicate on KFF-acpP. xanP, a nucleobase transporter, showed up as a private extra in the high-MIC M02 line and in a few other constructs. Those remain hypotheses, not rebuilt causes.

Peptide Choice Becomes a Clinical Design Filter

The field has spent years tuning oligo length, target site, and backbone chemistry. This ramp says the peptide is not a passive strap. KFF is cheap, familiar, and fragile in a way bacteria already know how to exploit. D-KFF and RXR look harder to train against in these four species, at least over 16 passages in broth. They are also chemically different beasts, with their own costs in synthesis, toxicity, and how well they work from one species to the next.

Our work shows that the delivery system is not just a carrier; it is a critical design feature that determines how readily resistance evolves.

Marco Galardini, TWINCORE group leader and senior author, Nature Communications

Galardini added that choosing delivery mechanisms less prone to resistance could improve how long programmable antibiotics last. That is a preclinical screen as much as a mechanistic finding: run the ramp on the peptide before you fall in love with the sequence. The paper is laboratory evolution, not a patient study, and it does not rank peptides for safety in people. It does say that a class advertised for speed of redesign can still lose if bacteria refuse the carrier.

No asobiotic has reached drug approval. A 2024 meeting at the Helmholtz Institute for RNA-based Infection Research in Würzburg even paused on the name, proposing “asobiotics” because antibacterial antisense oligomers still lacked a shared label. The open questions Vogel listed on his lab page, species specificity, off-targets, host response, and resistance, now have a resistance map for two peptides and four pathogens, which is narrower than a pipeline and more useful than another MIC chart.

Gram-Negative Infections Still Drive the Death Toll

The four species in the ramp sit inside the Gram-negative group that WHO and the Global Research on Antimicrobial Resistance project keep flagging as the thin stretch of the antibiotic pipeline. WHO’s fact sheet, updated 16 July 2026, says bacterial AMR was associated with more than 4.7 million deaths in 2021. The same page says about 1 in 6 laboratory-confirmed bacterial infections worldwide were resistant to antibiotics in 2023, and that resistance rose in over 40% of the watched pathogen-drug pairs between 2018 and 2023.

THE BURDEN THIS CLASS IS MEANT TO CUT

• Attributable deaths, 2021: The Lancet GRAM analysis put 1.14 million deaths directly down to bacterial AMR, inside the 4.71 million associated deaths.
• Forecast for 2050: The same project’s reference path reaches 1.91 million attributable deaths a year and, in an IHME newsroom summary, 39 million deaths from antimicrobial resistance in total from 2025 to 2050.
• Gram-negative gap: GRAM’s model said a working pipeline of new Gram-negative drugs could avert 11.1 million AMR deaths over that span.
• Money on the table: WHO cites a forecast of US$412 billion a year to treat resistant bacterial infections up to 2035, plus US$443 billion a year in lost productivity.

Carbapenem resistance in Gram-negatives was the fastest-rising class in that 1990-to-2021 window, and E. coli and K. pneumoniae lead drug-resistant bloodstream infections in the WHO write-up. An asobiotic that can be retargeted at acpP or rpsH, or at a resistance gene itself, is one answer people have wanted for that gap. The TWINCORE-HIRI ramp does not put such a drug on a ward. It says the peptide you pick will decide whether bacteria spend their mutations on a door you already know, or on a sequence you can still rewrite.

Frequently Asked Questions

What Are Asobiotics?

Asobiotics are antibacterial antisense oligomers, short synthetic nucleic-acid mimics written to bind a chosen bacterial RNA and stop an essential protein from being made. The name was proposed at an interdisciplinary meeting hosted by the Helmholtz Institute for RNA-based Infection Research in Würzburg on 12 and 13 September 2024, because the field still had no single label; that meeting also recorded that, despite years of in-vitro and animal proof, no antimicrobial ASO had reached drug approval.

Why Does the D-Form of the KFF Peptide Avoid the Usual Resistance Path?

SbmA moves PNA from the periplasm into the cytoplasm after the L-form (KFF)3K peptide has been cut by a periplasmic protease that has not been identified. The D-form of the same peptide resists that digest, so it does not depend on SbmA, and the E. coli lines in this study produced no sbmA mutations under D-KFF-acpP, with only a 1.3-fold mean MIC rise.

What Does the Bacterial prfB Gene Do?

prfB encodes peptide chain release factor RF2, which, among other jobs, works with ArfA to rescue ribosomes that have stalled. Mutations at amino acid 246 of RF2 were already known to raise termination efficiency, which is why the authors treated the T246S hit as a plausible, uptake-independent way to live with a drug that is supposed to stop the ribosome binding in the first place.

When Did Antisense Oligomers First Kill Bacteria?

Field reviews trace the first bactericidal antisense PNA to work that paired an oligomer against acpP, the acyl-carrier-protein gene, with the (KFF)3K peptide in E. coli, after earlier PNA experiments that blocked beta-lactamase and resensitized leaky strains to ampicillin. The first antisense phosphorodiamidate morpholino oligomer aimed at acpP was reported by Geller and colleagues in 2003, including a mouse peritonitis test that took the idea out of a test tube.

The paper is open access as Nature Communications volume 17, article 8262, received 14 January 2025, accepted 15 July 2026, and issued on 12 August 2026. The screen it offers is simple enough to copy: evolve the peptide first, then spend chemistry on the sequence.

Disclaimer: This article is news reporting and analysis of a laboratory evolution study, and it is for information only. It is not medical advice, is not a treatment recommendation, and does not tell anyone to start, stop, or switch an antibiotic or experimental antisense drug. Readers who have an infection, or who are considering a trial, should speak with a licensed physician or infectious-disease specialist before acting. Figures, gene names, and approval status reflect the cited sources as of 28 August 2026 and may change as further studies and regulators move.