Fig 1.
The developed system for the preprocessing, segmentation and tracking of epithelial cells.
A) Schematic of the computational pipeline from the acquisition of 3D time lapse data to image preprocessing, cell segmentation and tracking. After segmenting the cells, we define symbolically their structure using a planar graph connecting detected AJ vertices with edges (green). Then we identify the cells in the tissue as the faces of the AJ graph and build the Cell graph to describe cell connectivity (blue). Finally, we establish correspondence between cells among frames (colored lines connecting cell centroids) obtaining cell trajectories. (B-G) Part of an epithelium of a Drosophila leg at early pupal stages. This tissue dramatically narrows and elongates at this stage to generate a narrow and hollow cylinder while the epithelium at presumptive joints invaginates. B) Maximum intensity projection of an image stack through the leg epithelium marked with E-cad∷GFP to highlight cell outlines. Distal up, narrow region—presumptive joint; wider regions part of the presumptive segment. C) Projection of the denoised and deconvoluted volume. D) The output of the filters employed to detect AJs (green) and AJ vertices (red). E) AJ graph representing the AJs structure. F) Cell graph representing neighborhood relationships among cells in the tissue. G) Polygonal representation of the cells, colored according to assigned temporal identifiers.
Fig 2.
Detection of mitosis and apoptosis events.
In addition to cell movements, notum morphogenesis involves extensive cell delamination at the dorsal midline [30] and extensive cell proliferation required for notum expansion [7]. We therefore incorporated into the tracker algorithms to detect mitotic and apoptotic events. A) The large light green cell in the upper left region splits to produce two daughter cells marked by the same color. B) The dark brown cell in the center reduces its area until it disappears.
Fig 3.
Analysis of Drosophila notum morphogenesis.
A) A Maximum intensity projection of an image stack through the mid-scutum marked with E-cad∷GFP to highlight cell outlines. Anterior to the right. B) Maximum intensity projection of the output of the filters employed to detect the AJs (green) and AJ vertices (red). C) AJ graph (green) and Cell graph (blue) symbolically represent the cell outlines and their connectivity, respectively. D) Projection of the 3D centroid trajectories recovered after tracking the motion of cells. Note that motion of cell centroid varies depending on the position of the cell across the tissue. E) Evolution of cell strain rate parameters (, , , ) computed from Kalman smoothed trajectories of cell centroids over time. F) Mean velocities (, ) of the cell centroid trajectories over time. Time evolution of G) expansion coefficient ℰ, and H) rotation coefficient θ. See text for further detail.
Fig 4.
Performance assessment of the automated detection of vertices, cell-cell contacts and cell tracks.
A-D) Evaluation of the vertex detector, E-H) edge detector, and I-J) cell tracker. C-D), G-H) Green—true detections, blue—missed detections, red—false detections. A) Precision-Recall curve for AJ vertex detection. B) Variation of the F1 score of the vertex detector relative to changes in detection threshold TV. Vertex detection of C) Notum and D) Leg datasets. Vertex location accuracy highly depends on properly tuning up Tv. E) Precision-Recall curve for AJ edge detection. F) Variation of the F1 score relative to changes in propagation threshold TE. Edge detection in G) Notum and H) Leg datasets. Edge detection is more robust than vertex detection. I), J), K) and L) 2D projection of the trajectories respectively found for the cells in E) Notum, F) Leg, G) Mitosis in notum and H) Apoptosis in notum datasets. The system recovers accurate cell trajectories in different scenarios.
Fig 5.
Establishing connectivity among AJ vertices to segment epithelial tissues.
A) Voronoi regions are expanded from vertex locations to generate the Voronoi Diagram associating each voxel to the nearest AJ vertex. B) Supervertices are expanded inside Voronoi regions to link adjacent vertices. C) The AJs graph is built adding an edge between contiguous vertices through pairs of adjacent supervertices. D) Vertices, Voronoi regions and supervertices are superimposed.