38 of 39 Antibiotics Failed: Klebsiella in Fish Farms
Thirty-eight drugs out of thirty-nine did nothing. A 2026 study of farmed bullfrogs in central China ran a full susceptibility panel against a single isolate pulled from a sick animal, and exactly one compound came back with a susceptible reading. Polymyxin B, 17 mm zone. Ampicillin through meropenem, all resistant or intermediate.
The bug was Klebsiella pneumoniae. If you run a fish or amphibian operation, that name probably has never come up in a meeting. Which is sort of the point. Multidrug-resistant Klebsiella pneumoniae in aquaculture stopped being a story borrowed from human medicine the moment it turned up in a production pond, killed animals under controlled conditions, and showed up carrying the genetic equipment to do it again.
Why a hospital pathogen matters on a frog farm
In human healthcare, Klebsiella pneumoniae is one of the usual suspects behind difficult multidrug-resistant infections. Carbapenem-resistant lineages have been giving hospital formularies a beating for years.
What nobody had documented properly is what it does to aquatic animals. This 2026 work started because bullfrogs on a commercial farm began showing nervous signs. Heads held to one side. Stiff bodies. Poor coordination. Reddening along the abdomen and legs, and on post-mortem, swollen liver, spleen and kidney.
I keep coming back to the commercial angle here. American bullfrogs are high-value stock. A pathogen that produces neurological disease and then kills at scale does not trim your margin. It ends a cycle.
How the isolate was pinned down
The diagnostic route was conventional, which is worth laying out because any half-decent farm lab should be able to repeat it.
The dominant bacterium came out of a diseased bullfrog and got identified four ways at once: Gram staining, morphology, biochemical assays, and 16S rRNA sequencing. All four agreed. The isolate was then assigned to ST6839, a novel sequence type confirmed against the Klebsiella Pasteur MLST database. Its seven housekeeping alleles: gapA 44, infB 48, mdh 51, pgi 74, phoE 11, rpoB 68, tonB 39.
Typing is not paperwork. A sequence type is what lets you ask whether your bug is the same one that hit a farm three provinces over. A brand new type like ST6839 means the surveillance map is still being drawn.
The susceptibility panel, drug by drug
Testing used Kirby-Bauer disk diffusion, read against the 2025 standard disk diffusion breakpoints. These are measured zone diameters.
| Antibiotic | Disk load | Zone (mm) | Result |
|---|---|---|---|
| Polymyxin B | 300 IU | 17 | Susceptible |
| Nitrofurantoin | 300 µg | 16 | Intermediate |
| Compound sulfamethoxazole | 23.75 µg | 11 | Intermediate |
| Imipenem | 10 µg | 18 | Resistant |
| Meropenem | 10 µg | 16 | Resistant |
| Penicillin | 10 µg | 16 | Resistant |
| Carbenicillin | 100 µg | 12 | Resistant |
| Piperacillin | 100 µg | 6 | Resistant |
| Ampicillin | 10 µg | 9 | Resistant |
| Cephalothin | 30 µg | 12 | Resistant |
| Chloramphenicol | 30 µg | 10 | Resistant |
| Florfenicol | 30 µg | 13 | Resistant |
| Tetracycline | 30 µg | 14 | Resistant |
| Doxycycline | 30 µg | 13 | Resistant |
| Minocycline | 30 µg | 12 | Resistant |
| Gentamicin | 10 µg | 8 | Resistant |
| Trimethoprim | 5 µg | 10 | Resistant |
Two rows in there deserve a second look. The whole tetracycline class failed: tetracycline, doxycycline, minocycline, all resistant. That should bother anyone whose reflex response to a bacterial outbreak in fish or amphibians is a tetracycline.
Then there is imipenem. An 18 mm zone, bigger than polymyxin B, and still called resistant. A wide clear ring does not mean the drug works. Farmers read their own lab reports all the time and get exactly this wrong, and it costs animals.
And the single susceptible result across 39 drugs is not a treatment plan. It is a warning sign with a number attached.
What showed up in the genome
Sequencing ran on Illumina NovaSeq 6000, assembled with SPAdes v3.15.5 and polished with Pilon v2.2. One circular chromosome, 5,454,922 bp, 57.20 percent GC.
| Genomic feature | Value |
|---|---|
| Chromosome length | 5,454,922 bp |
| GC content | 57.20% |
| Predicted ORFs | 5,200 (mean 908 bp) |
| Coding fraction | 86.62% |
| tRNA genes | 63 (4,948 bp) |
| 16S / 23S rRNA | 1 (1,536 bp) / 1 (2,899 bp) |
| Genomic islands | 23 (4,287 to 50,061 bp) |
| CRISPR sequences | 3 |
| Antimicrobial resistance genes | 157 |
| Virulence-related genes | 179 |
Prediction against CARD returned 157 resistance genes. Efflux and target modification were the dominant mechanisms, and the determinants leaned toward cephalosporins, aminocoumarins, peptides and aminoglycosides.
VFDB annotation returned 179 virulence genes, among them 14 type VI secretion system genes, 13 T6SS-III associated genes, 11 capsule genes and 11 lipooligosaccharide genes. Named factors included mrkD, fimH, fimA, fimB, ompA, entC, entS, ecpB, gspG, impJ and iroE.
Twenty-three genomic islands, some of them tens of thousands of base pairs long. That is the fingerprint of a strain that has been shopping in its neighbours’ DNA. Resistance and virulence arriving together is not a coincidence. It is the same mechanism.
The challenge trial: how little bacteria it takes
Sixty healthy bullfrogs, 40 to 60 g, six groups of ten. Five groups got 200 µL of PBS by intramuscular injection carrying the isolate at 1×10^9, 1×10^8, 1×10^7, 1×10^6 or 1×10^5 CFU per animal. The sixth got plain PBS.
| Challenge dose (CFU/frog) | 14-day cumulative mortality |
|---|---|
| 1 x 10^9 | 100% |
| 1 x 10^8 | 100% |
| 1 x 10^7 | 100% |
| 1 x 10^6 | 60% |
| 1 x 10^5 | 50% |
| Control (PBS) | 0% |
LD50 by the Karber method came out at 2.51 x 10^3 CFU per bullfrog. That sits under the 1.0 x 10^5 CFU per animal line usually drawn for high virulence, so the isolate qualifies as highly virulent.
Log-rank (Mantel-Cox) comparisons: 1×10^7 versus 1×10^6, P = 0.0275. 1×10^7 versus 1×10^5, P = 0.0031. 1×10^8 versus 1×10^6, P = 0.0047. 1×10^8 versus 1×10^5, P = 0.0005. 1×10^9 versus 1×10^6, P = 0.0012. 1×10^9 versus 1×10^5, P = 0.0002.
The assumption worth dropping
There is a comforting piece of folk wisdom in this industry. Resistant bugs are weaker bugs. The thinking runs that a bacterium spending its energy on efflux pumps has less left for causing disease, so resistance buys you a bit of time.
This isolate will have none of it. 157 resistance genes, 179 virulence genes, 38 of 39 antibiotics useless, and it still killed every animal in three of five challenge groups with an LD50 in the low thousands of CFU. Nothing is trading off here. Both traits are stacked in the same organism.
The second assumption to bin is that Klebsiella is a human-medicine problem and aquaculture can sit this one out. Most antimicrobial resistance research runs farm to human, and fairly so. This case runs the other way: an organism shaped by clinical antibiotic pressure, turning up in an animal production system already tooled up to cause severe disease.
What multidrug-resistant Klebsiella pneumoniae in aquaculture means for your buying and treatment decisions
For producers, the sequence is dull and it works.
Stop defaulting to tetracyclines for an unidentified bacterial outbreak in fish or amphibians. This isolate beat tetracycline, doxycycline and minocycline, and florfenicol, another standby, failed too. If your protocol is tetracycline in the feed and a look-see in three days, you may be paying for selection pressure while the animals die anyway.
Run the susceptibility panel before you buy the treatment, not after. Disk diffusion is cheap, most regional labs offer it, and it reads against published breakpoints. Two days of waiting costs less than a lost cycle.
Treat neurological presentation in bullfrogs as an escalation trigger. Head tilt, stiffness, loss of coordination, swollen liver, spleen and kidney: that is not nutrition, and a water change will not fix it.
Then do the boring bits that cut pathogen load in the first place. Lower density. Quarantine new stock. Keep cohorts apart. Pull mortalities fast. Write down what you dosed and when. None of it is novel. It is what you have left when the drug list shrinks.
If you are buying aquatic protein or live stock, three questions will sort your suppliers. Do they test before treating, or treat on habit? Can they show mortality records by cohort? Who picks the antibiotic, a vet or someone standing at the pond edge? A supplier who answers all three on paper is a different risk from one who cannot.
Where this study falls short, honestly
One isolate, one farm, one location. That is a case report with a genome bolted on, not a prevalence survey. It says nothing about how common ST6839 is.
One strain was tested. No comparison panel of susceptible or moderately resistant isolates from the same system, so we cannot tell whether this one is unusually nasty or ordinary for what is circulating.
The challenge was intramuscular injection. Standard for establishing virulence and an LD50, but it skips the gills, gut and skin a natural infection has to get through. Field mortality at a given exposure could well be lower.
The gene counts come from database prediction, not expression work. A predicted gene is a hypothesis about phenotype. The plate is the phenotype. Where they disagree, believe the plate.
And no treatment trial ran at all. You learn what failed in vitro and how virulent the organism is in a challenge model. You do not learn what works in a pond, because nothing was tested for efficacy.
Frequently asked questions
How common is multidrug-resistant Klebsiella pneumoniae in aquaculture?
Nobody knows yet, and that is the straight answer. This work covered one isolate from one diseased bullfrog on one farm in central China. Enough to prove the organism can cause severe disease in an aquatic production species and to justify watching for it, but nowhere near enough to give a prevalence figure. Read it as a signal that the bug is present and dangerous, not as evidence of a widespread outbreak.
Why did only polymyxin B work against this isolate?
Polymyxin B gave a 17 mm zone, the only susceptible reading out of 39 antibiotics. The probable reason is that polymyxins attack the outer membrane directly, and this strain had apparently not picked up a defence against that yet. That is a thin advantage, and the class is one many regulators hold back for human medicine for exactly this reason. Read it as alarm, not comfort.
Is a large inhibition zone a reliable sign that an antibiotic will work?
No, and one row here settles it. Imipenem produced an 18 mm zone, wider than polymyxin B, and still read as resistant under the 2025 standard disk diffusion breakpoints. A zone diameter only means something against the published breakpoint for that drug and that organism. Read the interpretation table, not the ruler. It is one of the most common mistakes made when producers read their own lab reports.
Does antibiotic resistance make bacteria less virulent?
Not here. The isolate carries 157 predicted resistance genes and 179 predicted virulence genes, and it killed 100 percent of challenged animals at three dose levels. Its LD50 was 2.51 x 10^3 CFU per bullfrog, under the 1.0 x 10^5 line used to define high virulence. The fitness-cost argument did not hold. Betting stock on the idea that a resistant bug is weak will lose animals.
What clinical signs should bullfrog producers watch for?
Animals in the source case showed head tilt to one side, body stiffness, movement disorders and reddening on the abdomen and legs, with swollen liver, spleen and kidney internally. In the challenge trial mortality reached 100 percent within 14 days at 1 x 10^7 CFU per animal and above, while controls given sterile PBS showed none. Nervous signs in a batch should bring in a vet, not a bag of medicated feed.
What is sequence type ST6839 and why does it matter?
ST6839 is a multilocus sequence type defined by seven housekeeping alleles: gapA 44, infB 48, mdh 51, pgi 74, phoE 11, rpoB 68 and tonB 39. It was confirmed as novel against the Klebsiella Pasteur MLST database. Typing makes a bacterium trackable. Without it two farms cannot tell whether they face the same organism. That this one is new says the surveillance map is still close to blank.
Should tetracycline still be used in aquaculture?
Yes, with a lab result in hand. Tetracycline-class compounds still work against susceptible organisms, and oxytetracycline is a reasonable tool once susceptibility is confirmed. What this case shows is that blind use is a coin flip, since tetracycline, doxycycline and minocycline all failed at once. The order should be diagnosis, susceptibility testing, then treatment, not the other way round.
How many resistance genes did the isolate actually carry?
Whole genome prediction against CARD identified 157 antimicrobial resistance genes. Efflux and target modification were the dominant predicted mechanisms, and the classes skewed toward cephalosporins, aminocoumarins, peptides and aminoglycosides. These are homology-based predictions, so they describe genetic potential, not which mechanisms switch on during infection. Disk diffusion results are the evidence that counts.
How was the LD50 of 2.51 x 10^3 CFU calculated?
Sixty healthy bullfrogs of 40 to 60 g went into six groups of ten. Five received 200 uL of PBS by intramuscular injection carrying the isolate at doses from 1 x 10^5 to 1 x 10^9 CFU per animal, and the sixth received PBS alone. Cumulative 14-day mortality was 100, 100, 100, 60 and 50 percent across the dose groups and zero in controls. LD50 then came from the Karber method.
Can Klebsiella pneumoniae spread from aquaculture to people?
Careful here. Klebsiella pneumoniae is well established as a multidrug-resistant pathogen in human healthcare, and resistant bacteria and their genes do move between animals, people and the environment. This study tested no transmission, so no transmission claim comes out of it. What it does show is a highly resistant clinical-type organism now causing disease in farmed animals, a plausible route for exchange either way.
What biosecurity steps reduce risk on a bullfrog or fish farm?
The dull ones. Quarantine incoming stock before it meets the main population. Keep cohorts separate by age and do not drag equipment between them. Remove mortalities quickly and dispose of them away from the water. Keep density low enough that animals are not chronically stressed. Record every treatment with drug, dose, date and outcome. None of this makes an outbreak impossible, but together they cut the load it needs.
What should a buyer ask a supplier about antibiotic use?
Three questions sort the serious operations. Do you run susceptibility testing before treating, or treat on what worked last season? Can you produce mortality records by cohort? Who makes the antibiotic call, a vet or a pond-side manager? A supplier who answers all three with paperwork is far lower risk than one who cannot, and with resistant organisms turning up in new species, that paperwork is worth paying for.
