Brandon Niese (The University of Texas at Austin)

"Quantitative imaging analysis elucidates properties of complex biological systems"

Abstract: This study introduces innovative quantitative image analysis methods for studying complex biological systems. It explores three distinct systems: Pseudomonas aeruginosa biofilms in mouse wound models, the polyextremophile bacterium Deinococcus radiodurans, and the human skin microbiome. Regarding P. aeruginosa biofilms, the study investigates how varying levels of Extracellular Polysaccharide Substance (EPS) production impact the biofilm's spatial properties in mouse wound environments. Mutant P. aeruginosa strains lacking Pel and/or Psl exopolysaccharide production showed altered biofilm aggregate size and distribution in wound tissue, with reduced survival under antibiotic treatment, suggesting a potential role for Pel and Psl in bacterial persistence in vivo. In the case of D. radiodurans, the research examines changes in nucleoid shape in response to different radiation doses. A radiation phenotype is defined, and the study explores the effects of various protein knockouts and knockdowns on this phenotype. Results indicate that ionizing radiation (IR) induces cell redistributions across sub-populations, with some showing morphologies indicating increased nucleoid condensation and fewer cells involved in cell division. The role of nucleoid-associated proteins (NAPs) in nucleoid compaction regulation remains unclear. Imaging of genomic mutants lacking known and suspected NAPs suggests that certain nucleic acid binding proteins, not previously identified as NAPs, can alter nucleoid compaction even without stress, and IR exposure increases the occurrence of these changes. For the human skin microbiome, deep learning techniques are applied in image analysis workflows to predict taxonomic diversity and the most abundant bacteria class in a sample. The model demonstrates strong performance, achieving an MSE of 0.0321 +/- 0.0035 for predicting the Shannon Diversity index and an accuracy of 94.0% +/- 0.7% for predicting the top taxonomic class. These findings underscore the potential of deep learning in analyzing the human skin microbiome and suggest applicability to other bacterial microbiomes. Overall, this research showcases the effectiveness of quantitative image analysis in understanding key properties and dynamics of complex biological systems, offering insights that can guide further studies in microbiology and biomedicine.