Population genomics of Candida albicans reveals genetic footprints that allow the prediction of antifungal resistance

Abstract

Rapid detection of drug resistance in fungi is becoming increasingly urgent. While multiple loci and mutations were linked to antifungal resistance in Candida albicans, the utility of whole genome sequencing in predicting resistance has not been tested thoroughly. We address this by integrating more than 1000 C. albicans genomes, with associated resistance phenotypes to six antifungals, a curated repository of resistance-causing mutations and variant effect predictors. We used Genome Wide Association Studies and Machine Learning approaches to associate genetic variants to resistance phenotypes. We found that well curated databases like FungAMR are able to accurately identify markers predictive antifungal resistance. Curated mutations were detected in 56 to 74% of isolates with reduced azole susceptibility and 57 to 90% of those with reduced echinocandin susceptibility, revealing incomplete coverage of resistance across both drug classes. Complementary variant- and gene-level association analyses recovered established resistance determinants and uncovered additional candidate loci, whereas copy-number changes provided limited independent explanatory power. Genomic prediction declined when genetic backgrounds were absent from training, and when leading predictive features differed among backgrounds. We also found that models trained on a curated list of mutations and variant effect predictors performed well when classifying drug-susceptible strains, but the drug-resistance trait was predicted with lower accuracy. These findings support the potential of genome-based resistance prediction while highlighting the need for broader population sampling, functional validation of candidate variants and expanded resistance catalogues to improve prediction across genetic backgrounds.