EmbryoMiner: A new framework for interactive knowledge discovery in large-scale cell tracking data of developing embryos

Abstract

State-of-the-art light-sheet and confocal microscopes allow recording of entire embryos in 3D and over time (3D+t) for many hours. Fluorescently labeled structures can be segmented and tracked automatically in these terabyte-scale 3D+t images, resulting in thousands of cell migration trajectories that provide detailed insights to large-scale tissue reorganization at the cellular level. Here we present EmbryoMiner, a new interactive open-source framework suitable for in-depth analyses and comparisons of entire embryos, including an extensive set of trajectory features. Starting at the whole-embryo level, the framework can be used to iteratively focus on a region of interest within the embryo, to investigate and test specific trajectory-based hypotheses and to extract quantitative features from the isolated trajectories. Thus, the new framework provides a valuable new way to quantitatively compare corresponding anatomical regions in different embryos that were manually selected based on biological prior knowledge. As a proof of concept, we analyzed 3D+t light-sheet microscopy images of zebrafish embryos, showcasing potential user applications that can be performed using the new framework.

Fig 1. Different possibilities to visualize 3D+t cell tracking data.

(A) Temporally scrollable maximum intensity projection with superimposed centroids of detected cell nuclei (red dots) allows following development over time. (B) Similar to the maximum intensity projections, detected nuclei can be superimposed on interactive 3D volume rendering of the original data. (C) Tracks of all cells can be analyzed from arbitrary orientations in an interactive and highly responsive 3D visualization using plain colors (C1) or quantitative trajectory features (C2, color-code indicates time). The panels show neural crest cells of a zebrafish embryo at 20 hpf (A, B) and the trajectories spanning the entire experimental duration from 12.5 − 28 hpf (C). Scale bar: 100 μm.

Fig 2. Feature-based extraction of groups of interest.

(A) Quantitative features associated with each of the cell tracks allows feature-based selections. The left and right part of an embryo was separated along the anteroposterior axis using the end point coordinates of each track. (A1) and (A2) show the identified groups of ∼1400 neural crest cells as a scatter plot (x coordinate of the trajectory end points versus the unique track ID) and a 3D rendering of all trajectories. (B) Cluster algorithms can be used to automatically group the data. The example shows four identified clusters using the end point locations in the XY-plane at a selected time point as a 3D rendering (B1) and a scatter plot (B2). (C) Special trajectory features can be used to characterize particular cell movements in the early embryo. (C1) schematically illustrates the ratio of effective displacement (distance between start and end point) versus the spatial length (integrated path length) that was used to automatically identify two clusters corresponding to hypoblast cells (magenta) and epiblast cells (green) visualized as scatter plot (C2) and 3D rendering (C3). Panels (A) and (B) show neural crest cells of a zebrafish embryo (12.5 − 28 hpf) and panel (C) is based on a slice cut from a whole-embryo zebrafish data set (5 − 7.25 hpf). Scale bar: 100 μm.

Fig 3. Interactive selection possibilities complement the automatic feature-based group selections.

Tracking data from cranial neural crest cells of a zebrafish embryo during 14-20 hpf are shown. (A) All visualization windows allow selecting groups of interest using freehand selection tools. From the left: maximum intensity projection overlay of the raw images and cell centroids for two time points of the neural crest data set at 12.5 hpf and 14 hpf (A1, A2); tracks of all cells in 3D (A3); subset of selected tracks in 3D (A4); scatter plot of track-based features (A5). All visualization windows (A1-A5) are synchronized to obtain consistent selections in all views (see corresponding color-code of panels A1-A5). (B) Exemplary manual selection of a group of interest. Freehand selection tools allow intuitive and interactive selection/deselection of groups of interest. (C) All performed selection steps can be recorded in a hierarchical selection tree view and arbitrary nodes of the tree can be combined to new groups of interest. The hierarchical selection tree view serves as a template to reproduce a particular selection on other data sets including the possibility of refinements to adapt to biological variation of different data sets. The panels show neural crest cells of a zebrafish embryo (12.5 − 28 hpf). Scale bar: 100 μm.

Fig 4. Steps that were performed for extracting hypoblast cells in four different wild-type zebrafish embryos with developmental time ranging from 2–14 hpf (A).

The knowledge discovery process was designed interactively on Embryo 01. First, the embryo was filtered temporally (5–7.25 hpf) and spatially (region around the blastoderm margin) to focus on the region of interest (B, C). The two groups of cells were separated using a feature-based clustering approach (D-F). The whole analysis pipeline was then applied to Embryos 02-04 resulting in the extraction of the same internalizing cells in all embryos. The color code in panels (A-E) indicates time from 2–14 hpf and the group association to hypoblast (magenta) or epiblast (green) in panel (F).

Fig 5. Quantitative comparison of selected tissue deformation features recently published by [35] measured for hypoblast (blue) and epiblast (magenta) cells.

Each column contains the results obtained on one of four wild-type zebrafish embryos in a time interval spanning early gastrulation from 5 − 7.25 hpf. The selected features comprise speed, the rotation discriminant D, the volume change rate denoted by P and the distortion rate Q_d as described in the main text. Note that Embryo 03 physically moved during image acquisition, which caused the increased total speed. Despite this global speed difference, all other quantitative features were nicely captured and revealed comparable patterns among all analyzed embryos.

Fig 6. Interactive tracking correction of neural crest cells of a zebrafish embryo.

(A) To analyze the precursors of the olfactory epithelium, a particular subgroup of neural crest cells of a zebrafish embryo, a set of about 100 cells was interactively selected at 23 hpf (blue). The corrected tracks are shown in the global context (B) as well as in isolation using time for coloring (C). Scale bar: 100 μm.