Metabolic flux fingerprinting differentiates planktonic and biofilm states of Pseudomonas aeruginosa and Staphylococcus aureus

The challenges of defining the biofilm phenotype has been clear for decades. Many biomarkers for biofilm are known, but methods for identifying these are often invasive and/or complicated. These methods often rely on disrupting the biofilm matrix or examining virulence factors and compounds, which may only be expressed under certain conditions. We used microcalorimetric measurements of metabolic energy release to investigate whether unchallenged, planktonic Pseudomonas aeruginosa displayed differences in metabolism compared to surface-bound and non-attached biofilms. The pattern of energy release observed in the recorded microcalorimetric thermograms clearly depended on growth state, though the total energy expenditure was not different between growth states. To characterize these differences, we developed a classification pipeline utilizing machine learning algorithms to classify growth state, based on the observed patterns of energy release. With this approach, we could with high accuracy detect the growth form of blinded samples. To challenge the algorithm, we attempted to limit the amount of training data. By training the algorithm with only a single data point from each growth form, we obtained a mean accuracy of 90.5% using two principal components. Further validation of the classification pipeline showed that the approach was not limited to P. aeruginosa but could also be used for detection of gram-positive Staphylococcus aureus biofilm. We propose that microcalorimetric measurements, in combination with this new quantitative framework, can be used as a non-invasive biomarker to detect the presence of biofilm. These results could have a significant potential in clinical settings where the detection of biofilms in infections often means a different outcome and treatment regime for the patient.