Penicillin-Binding Protein Transpeptidase Signatures for Tracking and Predicting β-Lactam Resistance Levels in Streptococcus pneumoniae

ABSTRACT β-Lactam antibiotics are the drugs of choice to treat pneumococcal infections. The spread of β-lactam-resistant pneumococci is a major concern in choosing an effective therapy for patients. Systematically tracking β-lactam resistance could benefit disease surveillance. Here we developed a classification system in which a pneumococcal isolate is assigned to a “PBP type” based on sequence signatures in the transpeptidase domains (TPDs) of the three critical penicillin-binding proteins (PBPs), PBP1a, PBP2b, and PBP2x. We identified 307 unique PBP types from 2,528 invasive pneumococcal isolates, which had known MICs to six β-lactams based on broth microdilution. We found that increased β-lactam MICs strongly correlated with PBP types containing divergent TPD sequences. The PBP type explained 94 to 99% of variation in MICs both before and after accounting for genomic backgrounds defined by multilocus sequence typing, indicating that genomic backgrounds made little independent contribution to β-lactam MICs at the population level. We further developed and evaluated predictive models of MICs based on PBP type. Compared to microdilution MICs, MICs predicted by PBP type showed essential agreement (MICs agree within 1 dilution) of >98%, category agreement (interpretive results agree) of >94%, a major discrepancy (sensitive isolate predicted as resistant) rate of IMPORTANCE The human pathogen Streptococcus pneumoniae is a leading cause of morbidity and mortality worldwide. β-Lactam antibiotics such as penicillin and ceftriaxone are the drugs of choice to treat pneumococcal infections. Some pneumococcal strains have developed β-lactam resistance through altering their penicillin-binding proteins (PBPs) and have become a major concern in choosing effective patient therapy. To systematically track and predict β-lactam resistance, we obtained the sequence signatures of PBPs from a large collection of clinical pneumococcal isolates using whole-genome sequencing data and found that these “PBP types” were predictive of resistance levels. Our findings can benefit the current era of strain surveillance when whole-genome sequencing data often lacks detailed resistance information. Using PBP positions that we found are always substituted within highly resistant strains may lead to further refinements. Sequence-based predictions are accurate and may lead to the ability to extract critical resistance information from nonculturable clinical specimens.