Geometrical confinement controls the asymmetric patterning of brachyury in cultures of pluripotent cells

ABSTRACT Diffusible signals are known to orchestrate patterning during embryogenesis, yet diffusion is sensitive to noise. The fact that embryogenesis is remarkably robust suggests that additional layers of regulation reinforce patterning. Here, we demonstrate that geometrical confinement orchestrates the spatial organisation of initially randomly positioned subpopulations of spontaneously differentiating mouse embryonic stem cells. We use micropatterning in combination with pharmacological manipulations and quantitative imaging to dissociate the multiple effects of geometry. We show that the positioning of a pre-streak-like population marked by brachyury (T) is decoupled from the size of its population, and that breaking radial symmetry of patterns imposes polarised patterning. We provide evidence for a model in which the overall level of diffusible signals together with the history of the cell culture define the number of T+ cells, whereas geometrical constraints guide patterning in a multi-step process involving a differential response of the cells to multicellular spatial organisation. Our work provides a framework for investigating robustness of patterning and provides insights into how to guide symmetry-breaking events in aggregates of pluripotent cells.


Fig. S1 quantitative immunofluorescence and generation of the binned density spatial maps.
Micropatterning technique allows to force the cells to form colonies of a defined shape and size. This enables the possibility to determine the preferential distribution of a specific subpopulation within a given geometry using the following procedure: The cells are fixed and stained for both a marker of interest (here T is represented) and a nuclear marker (here Lamin B1) which is required to identify each individual cell (1). Imaging is performed using confocal microscopy* (2). A custom 3D nuclei segmentation method (manuscript in preparation and software available on request) allows for the automated identification of individual cells within each image (3). Once nuclei are segmented, the coordinates of the nucleus barycentre and the mean intensities of the signal in other image channels is computed. Also, the autofluorescence of the patterned substrate is segmented (3) in order to obtain the centre of the shape which is then used as a reference to normalise coordinates of the cells across multiple colonies. The cells are then classified using a threshold set manually based on the mean fluorescence intensity detected within the nucleus** (4). The binned density maps (BDM) are generated with ggplot2 using coordinates of cells accumulated over all the colonies imaged over all independent experiments (5). NB: The 'density' scale bar shown in the BDMs represent the frequency density of the 2D histogram, in other words, this represents the number of events found within a specific bin divided by the total number of event in the data.
* Due to the time required to image colonies at a resolution sufficient for accurate cell segmentation (~ 3 min / colony for ellipse M, 15 min for flowers), some form of colony sampling is required. To account for variations in background and possible staining inhomogeneities, we sampled colonies at various locations on the coverslip based on the nuclear signal (without looking at the T signal to prevent selection bias) and based on whether the colony morphology nicely followed the pattern shape visible by autofluorescence (the staining procedure sometimes led to colonies detaching from the pattern or to colonies with a sheared structure). The number of colonies imaged per condition largely depended on the number of undamaged colonies as well as time considerations. ** Selecting a threshold to define cells as positive or negative for a specific marker always contains a part of subjectivity (even when using a statistical method such as fitting a gaussian mixture as the choice of the method also requires a number of assumptions).
To select thresholds as objectively as possible, we created a scatter plot of the distribution of intensities for all the cells in the experiment independently of the culture condition. Our software allows for clicking on individual data points to visualise a thumbnail image of the corresponding cell. This permitted to iteratively refine the threshold based on both the shape of the distribution of intensities as well as based on the image visualisation while remaining blind to the sample under scrutiny for each click. Development: doi:10.1242/dev.166025: Supplementary information  The upper table summarises the properties of each geometrical design used in this study, including the edge to edge distance between shapes in both the x and y axis regardless of the geometry (pitch), the surface area of one shape, the total adhesive surface for one slide, the number of shapes on 1cm 2 chip and the percentage that this area represents compared to the area of a fully adhesive chip. Below the table, shapes are drawn and dimensions are indicated in µm. Development: doi:10.1242/dev.166025: Supplementary information

Examples of cells distributions within individual ESC colonies grown on discs or ellipses micropatterns. 3D
coordinates of the cells barycenters are projected on the XY plane and represented as circles. The blue line indicates the boundary of the micropattern. The percentage of T+ cells and the total number of cells (n Cells) is indicated for each plot. NB: All individual colony plots may be generated using the provided data text files and R code.

Fig. S4 Local cell density distributions
Representative heatmaps of the distribution of local cell density computed in a circular region of 75 µm of radius around each cell for ESC cultured at low, medium and high density or grown on disc or ellipse micropatterns. Densities range from 0 to 100 % of the maximum neighbour count identified in the image. T+ cells are shown as bright red cells and cells that are excluded from the neighbourhood analysis due to their proximity with the image border are shown in white. Scale bar: 200µm for low, medium and high densities, 100µm for Disc and Ellipse micropatterns.