Metagenomics approach to predict antibiotic resistance genes in sputum samples of adult people with cystic fibrosis: a pilot study: a pilot study

Lung infections in people with cystic fibrosis (CF) cause lung damage, which is the leading factor in the morbidity and mortality of CF. Prescription of antibiotics to treat these infections is essential to maintain a higher quality of life and increase life expectancy. Determination of antibiotic susceptibility (ABS) is done by culture-dependent, phenotypic methods. These procedures take several days, while timely intervention is key. The analysis of antibiotic resistance genes by use of shotgun metagenomics might offer a time-sensitive alternative. Twenty people with CF with a homozygous Phe508del mutation provided 68 sputum samples during different visits over a period of roughly a year. After shotgun sequencing, the samples were analyzed using the deep learning tool deepARG. These results were compared with the results from routine ABS testing. The performance was determined by area under the curve-receiver operating characteristics (AUCROC) and sensitivity. Significant results were obtained for the following antibiotic classes: aminoglycoside (AUCROC = 0.81 [95% CI: 0.67–0.95, sensitivity = 73%]), cephalosporin (AUCROC = 0.70 [95% CI: 0.54–0.86, sensitivity = 95%]), and fluoroquinolone (AUCROC = 0.73 [95% CI: 0.56–0.89, sensitivity = 88%]). For other antibiotic classes, results were not significant. Using antibiotic class-specific cut-offs for positive reads of ARGs, a metagenomics approach potentially offers a culture-independent and more time-efficient manner to predict ABS for commonly prescribed antibiotic classes for sputum samples of adult people with CF. The use of metagenomics and artificial intelligence in clinical care will add to more personalized care for people with CF as well as better antibiotic stewardship.