Fig 1.
Workflow to extract and define myocardial motion and deformation patterns during early heart morphogenesis.
A comprehensive pipeline for the 3D imaging of early heart morphogenesis. This figure provides an overview of the workflow used in this study. For full details of each methodological step, please refer to the MATERIALS AND METHODS section. The computational workflow comprises four main components: 1) Estimating Individual Live Image Motion (blue shapes): Following image preprocessing, we extracted the underlying motion using a non-rigid registration algorithm, resulting in a set of Transformation. The accuracy of motion detection was evaluated by comparing tracked cells to computed tracking using a stepwise and sequential interpolation. The myocardium tissue was segmented at one time point, and a mesh was generated. By interpolating this mesh with the set of T-transformation, we derived ‘Live-Shape’, a continuous description of heart tissue motion. 2) Integrating Multiple Live Images into the Atlas (red shapes): Individual live image motions were integrated into a high-resolution Atlas. This involved a staging system to synchronize ‘Live-Shape’ sequences along the Atlas time reference and a spatial mapping strategy to project staged ‘Live-Shape’s into the Atlas spatial framework. The result was a set of SurfaceMap, representing the motion of each specimen within the Atlas. 3) Quantifying Tissue Deformation (yellow shapes): Individual tissue deformation patterns were extracted, mapped into the Atlas, and variability was assessed. We defined stepwise tissue deformation and cumulative deformation to quantify these changes over time. 4) Creating an In-Silico Fate Map (green shapes): An in-silico fate map of the myocardium was constructed for the developmental window between E7.75 and E8.25 by concatenating motion profiles, providing insights into the spatial and temporal dynamics of early heart morphogenesis. Our dataset includes multiple specimens raging from E7.75 to E8.25 (12 hours). Two different Rosa26Rtdtomato+/-, Nkx2.5eGFP specimens are aligned on a pseudo-timeline, representing the transition from the cardiac crescent to the linear heart tube.
Fig 2.
Estimating individual live image motion to describe heart tube shaping as a continuous process.
(A), MIRT algorithm was applied frame-by-frame (I(ti)-I(ti + 1)), extracting the set of deformation fields (T1, T2, …, TN) underlying the live images. (B), The top image illustrates the automatic cell tracking (blue tracks) on a live image (Nkx2.5eGFP; Rosa26Rtdtomato + /-). The bottom image is an Imaris reconstruction of the HT and the relative automatic cell tracking. This tracking serves as the ground truth dataset. On the left, there is a representation of the validation strategy. By applying the deformation fields, we built our test set (pink lines). We computed the Eulerian distances between the ground truth and the test set to assess the accuracy of the registration method. The scale bar is 100 µm and the time is represented in [hh:mm]. (C), The graph illustrates the method for generating the stepwise test set. For each centroid in the ground truth dataset, the T-transformation determines its new position (x′, y′, z′) at the next time point. (C’), The sequential test set is generated by sequentially interpolating the initial cell positions (x, y, z, t0) (shown in blue) using the transformation set {Ti}. The output from each iteration (shown in pink) serves as the input for the next iteration. (D), Average stepwise error in µm for a single embryo (e16) as a proof of concept. The black line represents the average stepwise error for 58 cells tracked over 204 minutes (35 frames), while the accumulated error is shown in pink. (E), Bidirectional approach started non-rigid registration at t(N/2), N = number of frames. (E’), Quantification of the accumulated error when the initial registration is fixed at t(N\2). (F), The comparison of average error and computation time trends of images rescaled at 10, 15, 20 and 25% of 3 embryos (e01, e02, e05). The analysis refers to the stepwise registration between only two consecutive frames. Values m1 and m2 represent the slope of the linear regressions of the two trends. (F’), The graph shows the percentage of points for which error between predicted and actual positions is greater than the mean diameter of a cell (d = 20 μm) for different rescaling rates (colour code). The dotted line represents the threshold we set to choose the resolution that best guarantees a balance between registration error and computation time. The threshold is set at 180 min of registration time. (G), Cardiomyocytes diameter is determined by calculating the maximum Feret diameter from the 3D segmentation of 26 cells. (H), The images provide a close-up view of cells under two different conditions: one where cells had already divided at time 0 (blue spots), and another where cells had not yet divided at time 0 (green spots). The orange mask outlines the segmented cell tissue at time 0 and after applying the T-transformation (t-2h; t + 2h). Dotted lines illustrate the actual elongation of the tissue at t-2h and t + 2h, while the straight line within the tissue indicates the direction of maximum elongation of the transformed segmentations. (H’), Cosine similarity was calculated between the actual tissue elongation and the elongation predicted by the T-transformation for 8 cells within the -2h to +2h interval. (I), 3D image of the HT of an Nkx2.5eGFP; Rosa26Rtdtomato + /- embryo(e02). The image includes a single ventral plane in a pink frame, and a single lateral plane with its segmentation section in a magenta frame. The myocardium (red), splanchnic mesoderm (green), and endoderm (blue) are visible in the segmentation section. The yellow arrow indicates the incomplete part of the right IFT. The blue arrow indicated the signal degradation in OFT. The segmentation was transformed into a volumetric mesh, defined as faces (yellow area), nodes (blue dots) and edges (orange sides). Scale bar: 100 µm. (L), Sequential interpolation between the mesh nodes with the deformation set returns a continuous tracking of the HT tissue (illustration related to e02). The points are random taken on the surface; they move in space following the tissue morphogenesis. The colour map is related to the position of the spots every hour. Yellow represents the position of the nodes at the initial time, orange after 1h, pink after 2h and purple after 3h. The grey line keeps track of the node trajectory. (M), Embryo during gastrulation. The tracking follows 25 cells for approximately 170 min. The arrows in green indicate the direction of cell motion, while the blue line tracks the entire path. (b), On the top, evaluation of the continuous error in 25 cells tracked during gastrulation at each 20 min interval. The dashed line indicates the slope (mG = 3.84) of the line that fits the median error values at each interval. Scale bar: 15 µm; video time resolution: 10 min; duration of the video: 170 min. (M’), On the bottom, error related to cell tracking during HT morphogenesis at each 20 min interval. The results are relative to tracking 30 cells during 216 min. The line fitting the median values of the intervals has a slope equal to mHT = 0.84. (M”), Evaluation of the tracking error normalised for the respective cell displacements. In blue, mean values and standard deviation are shown for mesodermal cell tracking during gastrulation at 20 min intervals. In grey, values related to HT cells.
Fig 3.
Integrating multiple live images into a consensus spatiotemporal reference by aligning each continuous motion profile with a published 3D + t Atlas.
(A), The static Atlas is described by 10 representative heart shapes. On the right, the morphometric parameter d1/d2 is shown. (B), Proposed features are defined as the Eulerian distances h, w, s, and θ. (B’), From left to right, a linear regression profile is shown for the ratios h/w, h/s, and θ versus d1/d2. Each scatter plot displays all Atlas specimens belonging to groups 1 to 9 (48 out of the total 50 specimens). The different groups are represented by a colour code. For each scatter plot, the coefficient of determination (R²) is reported. The h/w ratio has the highest correlation value with d1/d2, with R² = 0.85, while the linear regression models for h/s and θ returned R² values of 0.49 and 0.52, respectively. (C), Staging System: A Gaussian Mixture Model (GMMs) associates to the Atlas groups (indicated by colour code) the live image frame with highest probability (denoted by stars). (D), Results of the staging system. (E), TMM rigid registration rotates, translates, and resizes the Atlas shape (blue surface) until it overlaps with the ‘Live-Shape’ (grey surface). Parts of the Atlas shape corresponding to the missing IFTs and OFT in the ‘Live-Shape’ (indicated by red arrows) are removed, resulting in the Atlascut. (E’), Image non-registration (MIRT) is used to transform the ‘Live-Shape’ mask into the Atlascut mask. (E’‘), The T-transformation adjusts the position of the ‘Live-Shape’ nodes (grey point-cloud) to fit the Atlascut (blue point-cloud) morphology. The morphed shape is called SurfaceMap. The coloured points indicate the same nodes in their configuration before and after the transformation. Face-to-face matching between the Atlas shape and the ‘Live-Shape’ is performed between the centroids of the faces (marked by stars). (F), Validation of the spatial mapping is conducted by computing the mesh area in the ‘Live-Shape’ and plotting the growth values onto the Atlas face-to-face.
Fig 4.
Quantifying tissue deformation during early morphogenesis.
(A), The triangles of the mesh were transformed according to the T-transformation from ti to ti + 1. Colours code indicated the same triangles in their rest and deformed state (shapes refer to e27). (A’), The tissue deformation pattern into is plotted in ‘Live-Shape’(ti + 1) and mapped face-to-face into the synchronise Atlas group. We compute the individual deformation pattern for each staged ‘Live-Shape’ and for each Gr. (B), Schematic overview of the pipeline for concatenating multiple motion profiles using the EYE-PC and HOOK-PC strategies. SurfaceMap e27 at Grn represents the EYE-PC (red shape and dots). To predict the position of SurfaceMap e27 in subsequent groups (Grn + 1, Grn + 2), we used known SurfaceMap from another specimen, referred to as the Hook-PC (blue shape and spots). The closest points in the HOOK-PC to the EYE-PC at Grn were selected as corresponding points (blue rhombuses). The position of SurfaceMap e27 (i1, i2,..., in) in Grn + 1 and Grn + 2 was then determined by the positions of the selected points from the Hook-PC. (C), Live images concatenation path. Embryo e27 is the reference. In dark grey the eye-PCs are highlighted, in light grey the hook-PCs. (D), The cumulative anisotropy rate (θ_)of each mesh face (i) between two arbitrary Grs is given by the product of all the intermediate θ_gr. (E), Stepwise vs Cumulative anisotropy (θ). Cumulative deformation is computed from Gr2. Colour bars indicate the deformation magnitude. θ = 1 isotropic deformation. θ_ > 1anisotropic state. Caudal view of the heart tube is shown.
Fig 5.
Creating an in-silico fate map to analyse heart tube morphogenesis at both cellular and regional levels.
(A), Concatenation path of the individual live image. We concatenated through the hookPC and eye-PC strategy the videos related to embryos e31, e16, e35, e27, e24 and e12. The reference embryo is e35. (B), In-silico fate map tracking early cardiac morphogenesis as a continuum. The tool in Imaris allows to select points or regions in the myocardium and follow their motion over time from Gr2 to Gr9 in Atlas. In grey the point-cloud, in yellow the selected points track the regional changes at different stages. The lines, on the other hand, draw the displacement of the point-cloud. (C), The selected 10 arbitrary zones of the myocardium on the Gr9 shape. (C’), The displacement of the 10 zones in the different groups of the Dynamic Atlas. The two rows above represent the anterior part of the heart, the two rows below the posterior part. (D), Strategy to identify the same 10 anatomical zones in the live embryo. For each group, ‘Live-Shape’s were rescaled with a rigid registration algorithm to the Corresponding Atlas shape. Then the area of each mesh was calculated (colour map) for each rescaled ‘Live-Shape’s. Finally, the area values were averaged within the Atlas reference and summed for each of the 10 areas. (E-E’), The ‘Live-Shape’ and the Dynamic Atlas growth profiles of the 10 zones, interpolated for each Grs with a 3degree B-Spline. (E”) The validation strategy. (F) Top: A selected region in the Dynamic Atlas, illustrating how the region deformed anisotropically over time. Bottom: Cumulative anisotropy from Gr3 to Gr9. (G), At the cellular level (left), we tracked cell positions in e02 at Gr3 and Gr4 using Imaris (pink spots). On the right, we marked the initial positions of these cells on the fate map at Gr3 and then evaluated their predicted final positions at Gr4 (pink spots). (G’), At the regional level, we tracked different groups of points on the myocardium mesh of e27 from Gr4 to Gr5 (pink and yellow points). On the right, we identified the same groups of points on the fate map at Gr4 and assessed where these points ended up in Gr5. (H), Cellular-level error across 40 tracked cells from e01, e02, e05, and e16, compared to the corresponding predicted positions from the dynamic atlas. On average, the error is 20 µm, roughly equivalent to a cell diameter. Only 2 cells show an error greater than twice the cell diameter. We consider 95% of the cells to be well-predicted.