A landmark-free morphometric staging system for the mouse limb bud

Abstract

We have created a 2D morphometric analysis of the developing mouse hindlimb bud. This analysis has provided two useful resources for the study of limb development. First, a temporally accurate numerical description of shape changes during normal mouse limb development. Second, a web-based morphometric staging system, which has the advantage of being easy to use, and with a reproducibility of about ±2 hours. It allows users to upload a dorsal-view photo of a limb bud, draw a spline curve and thereby stage the bud within a couple of minutes. We describe how the system is constructed, its robustness to user variation and illustrate one application: the accurate tracking of spatiotemporal dynamics of gene expression patterns.

Fig. 1.

Current staging approaches for limb development. (A) A selection of right hindlimb buds photographed from the dorsal side to illustrate the sequence of normal limb development. (B,C) The high degree of size variability in embryo size is shown by linear measurements of footplate width (B) and crown-rump length (C) for 663 embryos harvested at six fixed timepoints after mating. Each embryo is represented as one datapoint and the huge degree of overlap between adjacent 12-hour timepoints is evident. (D) Five limb buds from embryos all staged as TS19 illustrate the low temporal precision of the Theiler staging system (Theiler, 1972). (E) A comparison of relevant staging systems highlights their strengths and weaknesses. Both Theiler stages and somite stages vary in length during development (becoming slower at later timepoints). The Wanek system (Wanek, 1989) is limb specific and therefore can stage fore- and hindlimbs independently, but has low temporal resolution. A rough alignment with the chick Hamilton and Hamburger system (Hamburger and Hamilton, 1992) is also displayed for comparison.

Fig. 2.

Quantifying limb bud shape over time. (A) The outline of a young limb was determined using a spline curve (yellow line). (B) The curvature graph derived from this spline. (C,D) Multiple shapes of a similar age can be compared to each other by aligning their curvature graphs (C), extracted from their photos (D). (E-L) The same set of splines and alignments is shown for two older limb buds. Wrist regions can always be seen as negative curvatures (concavities) to the left of the vertical origin. Digits produce positive curvatures (convexities) to the right of the vertical origin. (M) The full interpolated series of curvature graphs is shown, with the primary timepoints in black. (N) Any of the curvature graphs can be converted back into a 2D limb bud shape, creating the full trajectory of normal developmental shapes. (O) The standard trajectory can be visualized in 3D view, the convex regions such as digits (red) and concave regions (blue) highlight the dramatic shape changes over 36 hours of development.

Fig. 3.

The morphometric staging system. (A) The live screenshot show the GUI a user operates in order to stage a limb bud (publicly available at http://limbstaging.crg.es). The applet offers control over brightness, contrast and magnification, and the image can freely be moved in the window. The spline is drawn in the ‘drawing mode’ by clicking on the outline of the limb bud. A control point is added and the control points are connected by a spline interpolation. Additional data about the limb (such as strain, gastrulational age, its position and the pixel size of the microscope image) can be added and are stored in a database. The green ‘stage’ button opens a new window/tab with the staging result. (B) Graph illustrating the robustness of the system to limb orientation during photography. Within a range of 30°, the result varies by just ±1 hour. (C) Graph illustrating the robustness to user-dependent variations from five novice users. y-axis labels are 12 hours apart, and the average s.d. across the 12 limb buds was ~45 minutes. (D) The influence of tissue processing on the staging results. Sixteen specimens of ~mE11:09 were dissected, fixed, dehydrated and rehydrated. For each step, a photo was taken and staged. Results are shown for two versions of the staging algorithm. In pale colours are the size-dependant results that show a strong dip in predicted stage upon dehydration, whereas in bolder colours are the results of the size-independent algorithm. (E) The training dataset was analysed for left-right asymmetry. Size and colour show the number of limb pairs staged the same. There is an average asymmetry of nearly 2 hours, but no strong bias to one side.

Fig. 4.

The dynamics of Sox9 expression. (A) A series of six left-right pairs of Sox9 patterns all taken from a single litter. The youngest limb buds are roughly 24 hours younger than the oldest buds. By ordering the limb buds based on their morphometric stage, we see the progression of a subtle aspect of the Sox9 pattern. Strikingly, in the fourth pair of buds, the left bud is staged 4 hours older than the right (mE12:12 versus mE12:08) and the Sox9 pattern is slightly more developed (the expression domain connecting digits 1 and 2 is weaker, which is more similar to the older patterns of the litter). (B-D) A comparison of Sox9 progression between three different strains of mouse: (B) CD1, (C) OF1 and (D) C57Bl6. Although individual variations exist in the subtle details of the pattern, an agreement between pattern and morphological stage is seen across the three timepoints that were roughly 6 hours apart. Two limb buds are shown for the OF1s and CD1s of each stage to highlight the slight individual-to-individual variation. Further details of the full Sox9 time-course are provided in Fig. S4 in the supplementary material.