Rapid Bacterial Identification and Antibiotic Susceptibility Testing through Interferometry-based Surface Topography Measurement
Scientists have long classified organisms based on shared morphological characteristics. However, modern taxonomy, particularly for bacteria, is largely defined at the genetic level. Yet the information encoded in genetics propagates through many levels of biological organization: genes determine patterns of gene expression and protein production which, in turn, influence cellular physiology and behavior which, in turn, influence the collective morphology of growing populations. It is unclear how far up this hierarchy that the lower level taxonomic information remains preserved. Here, we show that bacterial genera can be identified from differences in the three-dimensional topographies of their growing populations. Using coherence-scanning white-light interferometry (WLI), which is capable of nanometer-resolution, we measure bacterial population topographies of 77 different clinical isolates representing four different genera. We then extract ten biophysically relevant features describing their surface structure. Using a simple machine-learning classifier, these topographic features identified bacterial genus with 97% accuracy, demonstrating that genus-level information remains present at the scale of population morphology. We hypothesize that these topographic "fingerprints" arise because differences in the underlying biophysics of bacterial growth propagate upward into measurable differences in population structure. To test this hypothesis, we performed biophysical simulations of bacterial population growth and applied our experimentally trained classifier directly to the resulting simulated topographies. By systematically modifying simulation parameters, we determined which changes were necessary to generate topographies classified as each of the experimentally observed genera. We find that a relatively small number of differences in cellular growth, cell-surface interactions, and initial conditions are sufficient to recreate the genus-associated differences observed experimentally, and we then show that using the topographic features from the simulations we can recover the underlying parameters used to simulate the topographies. Together, these results show that differences in lower-level bacterial behavior can propagate into distinct population-scale morphologies, and that taxonomic information can remain recoverable far above the molecular scale at which bacterial identity is conventionally defined.