Fusing proteins to fluorophores is standard procedure to make them visible under the microscope. By following these fusions in time, we can determine the movement and interaction patterns of interesting molecules inside the cells. Our standard techniques offer a well-resolved image of the movement in 2D but are blind to translocations along the z-axis. In this project, we use deep learning to accurately locate fluorophores in 3D inside living Escherichia coli.

To train the network, we simulate ground truth training data based on background models generated from experimental data. We test the method by studying how chromosomal loci are relocated in 3D over the E. coli cell cycle. Since the cells are radially symmetric, we expect the distribution of the fluorophores in y and z to look the same, and this is also what we observe for all the labeled loci. Interestingly, some loci are located exclusively in the periphery of the nucleoid, while others are more confined to the core of the nucleoid. This pattern would not have been visible had we tracked the loci only in 2D. If we combine the algorithm with microfluidic chips to study many genetically diverse strains simultaneously, we have a great tool to study the 3D chromosome dynamics at high spatial and temporal resolution.

Read more in Communications Biology!