What is the problem?

Bloodstream infection results in sepsis, which kills more than 40,000 people in the UK every year. Quickly providing a working antibiotic is necessary to reduce morbidity and mortality. However, currently, it takes 24-48 hours to grow bacteria from patients’ blood, and then another 18 h to determine antibiotic susceptibility. This means that patients can be prescribed non-functional antibiotics, or extremely broad spectrum, “last resort” antibiotics – just in case of resistance – upon initial diagnosis.

What is the solution?

It is possible to sequence the genomes of bacteria in blood samples without significant culture. Since all antibiotic resistance (ABR) is encoded in the genome sequence, it is therefore theoretically possible to predict whether the bacterium causing an infection is resistant to antibiotics because it carries certain genes. However, our work has demonstrated that ABR is often complex and involves particular levels of gene expression, not only whether a gene is present or not. Proteomics is a methodology that allows quantification and identification of proteins, so, since gene expression leads to protein production, it can give information about what genes are being expressed and to what extent. We have pioneered the use LC-MS-MS proteomics to help better predict antibiotic resistance using whole genome sequencing, by uncovering the biology of resistance and bridging the gap between genotype and phenotype.

Even if ABR can only be predicted following positive blood culture, this would still save 18 hours over current culture and susceptibility testing protocols. MALDI-TOF is another approach that can be used in proteomics. But in clinical diagnostics, it is used to generate a spectrum following the firing of a laser onto bacterial samples. The spectrum represents proteins and other substances in the bacterial envelope. This is standard practice for identifying the species of bacterium causing bloodstream infection, and can be applied directly using culture positive blood. We are trialling to see whether samples used for MALDI-TOF species identification can also be used in LC-MS-MS proteomics to identify and quantify antibiotic resistance proteins, and so predict resistance. Furthermore, we are machine-learning and AI approaches to test whether MALDI-TOF spectra, generated during routine diagnostics to identify bacterial species, can also be used to predict antibiotic resistance. This would be quicker and less expensive than applying LC-MS-MS.

Outcome and next steps 

We have successfully identified and quantified bacterial proteins in blood samples from septic patients and used this to correctly predict resistance to beta-lactam antibiotics, the most commonly prescribed antibiotics in humans (Takebayashi et al, 2022 BioRxiv 2022.02.27.482154). We have used proteomics to understand the complex interplay of factors that leads to fluoroquinolone resistance, and this has led to the establishment of 47 rules that predict fluoroquinolone resistance directly from whole genome sequencing (Wan nur Ismah et al, 2018 Antimicrobial Agents and Chemotherapy 62:e01814-17). We have identified how gene dosing and expression affects antibiotic susceptibility test outcome for critically important antibiotics and shown how proteomics can be used to predict aminoglycoside resistance, and unravel the complex interplay of factors that affect beta-lactam/beta-lactamase inhibitor resistance (Dulyayangkul et al., 2024 PLOS Pathogens 20:e1012235). We have also started developing bioinformatics pipelines that factor in gene expression and so improve prediction of antibiotic resistance from whole genome sequence (Reding et al, 2024 Briefings in Bioinformatics 25:bbae057).