Resistance does not follow the label: dose and background rewrite antifungal predictions
Three preprints point the same way: yeast evolution depends on dose and combination, Candida genomes expose gaps in resistance prediction, and an antibiotic worsened flu outcomes in the reported study.
- In a preprint, yeast genomic responses to antifungal selection were non-additive and depended on drug concentration and combination. Self-reported m:…1678
- In a preprint, curated databases detected resistance mutations in 56 to 74 percent of azole-resistant and 57 to 90 percent of echinocandin-resistant Candida albicans isolates. Self-reported m:…0196
- In a preprint, amoxicillin given during influenza A infection increased mortality and the severity of secondary bacterial pneumonia, with a more pronounced effect in males. Self-reported m:…2416
- In a preprint, evolved antifungal resistance in yeast showed little evidence of fitness costs in drug-free media, so resistant strains did not simply fade without the drug. Self-reported m:…1678
The dose changes the answer
In one preprint, researchers evolved yeast under antifungal selection and found that genomic responses were non-additive and depended on drug concentration and combination, which makes resistance harder to predict from a single drug or a single dose. m:…1678
The same preprint reports a consistent target alongside that variability. FKS1 was a consistent target for caspofungin resistance across varying concentrations, while high-dose clotrimazole selection frequently led to gain of the ERG11-containing chromosome VIII in the yeast species studied. In other words, one resistance route stayed stable while another depended on how much drug was used. m:…1678
The preprint also reports that evolved resistance showed little evidence of fitness costs in drug-free media. That detail matters because it weakens the comforting assumption that resistant organisms will automatically lose out once treatment stops, at least in the experimental conditions reported. m:…1678
Genomes predict susceptibility better than resistance
A second preprint, based on large-scale genomic analysis of Candida albicans, finds a similar prediction gap in a clinical pathogen. Curated databases such as FungAMR detected resistance-causing mutations in 56 to 74 percent of azole-resistant isolates and in 57 to 90 percent of echinocandin-resistant isolates, A detection rate in that range leaves a substantial share of resistant isolates unexplained by the listed mutations. m:…0196
Models built from curated mutation lists and variant effect predictors performed well at classifying drug-susceptible strains but did worse on drug-resistant phenotypes, according to the preprint. Prediction accuracy also declined when training data lacked representation of specific genetic backgrounds, so a model trained on one population may not travel to another. m:…0196
Taken together, the yeast and Candida preprints describe two different reasons for the same problem. In yeast, the selection regime itself reshapes the outcome. In Candida, incomplete coverage of resistance determinants and uneven genetic backgrounds limit what a genome alone can say about a resistant isolate. m:…1678 m:…0196
The wrong drug can harm, not just fail
A third preprint extends the caution from fungi to a common antibiotic used during viral infection. In the study, amoxicillin given during influenza A infection increased mortality and increased the severity of secondary bacterial pneumonia. The negative impact was more pronounced in males, according to the preprint. m:…2416
The finding challenges routine empiric antibiotic prescribing during viral respiratory infection, because the drug did not merely fail to help in the reported model. It was associated with worse outcomes, which points toward diagnostic-guided treatment rather than automatic coverage, pending further evidence. m:…2416
What this means for prediction
Across the three preprints, the practical lesson is narrow and consistent. Dose and combination matter in yeast evolution, genetic background matters in Candida prediction, and the infecting virus matters for an antibiotic given alongside it. A prescribing or surveillance decision that ignores those conditions is working with less information than the biology appears to require. m:…1678 m:…0196 m:…2416
The limits are narrower than a general warning. Candida prediction accuracy declined when training data lacked representation of specific genetic backgrounds, and models built from curated mutation lists and variant effect predictors did worse on drug-resistant phenotypes than on drug-susceptible strains. m:…1678 m:…0196 m:…2416
The three results also differ in what each study actually measures, which is why they point the same way without saying the same thing. The yeast study tracks genomic responses to selection under varying drug concentrations and combinations, the Candida study tracks how often curated databases detect resistance-causing mutations in resistant isolates, and the amoxicillin study tracks mortality and the severity of secondary bacterial pneumonia during influenza A infection. A prediction or prescribing rule that ignores those different measures will look precise while missing the condition that changes the outcome. m:…1678 m:…0196 m:…2416
What is still unresolved
- No functional validation for many Candida candidate variants identified in the genomic analysis.
- Whether resistance prediction improves enough for clinical use once more genetic backgrounds are represented in training data.
Provenance by Synorb. Every claim links to its source. This story was written by an agent from the Manifests listed below. Read the methodology and limitations, or send a correction.
Free to reuse, including commercially. Credit “Hanging Context, with provenance by Synorb” and link to this story. Keep the cited Manifest ids. Reuse terms v1.