PlantSeg-MorphoGraphX-3DCellAtlas workflow

For segmentation of root micrographs, the channel of the stained cell walls was extracted as a 16-bit.tiff image. The PlantSeg workflow (Wolny et al., 2020) was used to predict cell boundaries and label the cells in the image stacks. The re-scaling factor for the images was calculated based on their resolution and the model used for cell boundary prediction ([1.70, 2.8, 2.8] to fit root meristems to the ≪ confocal_unet_bce_dice_ds2x≫ model; and [1.74, 2.04, 2.04] to fit the mature root images to the ≪confocal_unet_bce_dice_ds3x≫ model). Graphics processing unit (GPU)-based prediction was used for cell boundary prediction, followed by segmentation using the generalized algorithm for signed graph partitioning (GASP) segmentation algorithm with beta: 0.5, 2D watershed, and a watershed threshold of 0.5. Segments smaller than 1000 were discarded. After segmentation, images were re-scaled with the appropriate factors and saved as .tiff image files.

Segmentations were imported into MorphoGraphX software (Barbier de Reuille et al., 2015). 3D meshes of each root were created using the marching cubes 3D workflow with a cube size of two and a minimal cell size of 500 voxels. Cell properties were extracted using the 3DCellAtlas plugin as described (Montenegro-Johnson et al., 2015). Cell types were assigned based on the radial and circumferential coordinate of the cells and manually corrected. Attributes and properties of all cells were exported for comparative analysis. Images of the labeled root meshes using horizontal or transverse cutting planes and of the original image were generated with MorphoGraphX. Videos of 3D image stacks and renderings were generated using Imaris software.