Creating digital atlases of animal morphology and gene expression is an emerging multidisciplinary field. The goal is to quantitate complex tissue structures and gene expression patterns as a digital record from which biological processes can be modeled and studied. Mitosis, apoptosis, differentiation and morphogenic movements result in complex embryo morphologies, and are driven by the expression of thousands of gene products that vary in amount from cell to cell throughout the embryo. Capturing such information requires high resolution embryo image data and sophisticated computational techniques to recognize thousands of cells, which vary from each other in size, shape and tissue type. Our goal is to create methods to produce a quantitative, cellular resolution map of gene expression and morphology for all of embryo development for Drosophila melanogaster. We have made significant progress towards this goal. First, for the early stage blastoderm embryo-a relatively simple, single layer structure of 6,000 cells surrounding a yolk-we created an extensive morphology and gene expression atlas that revealed previously unknown processes, such as nuclear movements. Second, funded by the previous round of this grant, we have initiated development an atlas for late stage embryos, which comprise 40,000 cells, ~70 tissue types and all major larval organs. The late embryo is significantly more complex that the blastoderm and presented new challenges: higher resolution images were required to resolve the more densely packed cells, and it proved essential to assign each nucleus/cell to a specific tissue type prior t nuclear/cell identification. This last task was difficult because it is only practical to label a gven sample with a few specific probes physically, yet ~70 different labels are required to identify each tissue separately. To meet this challenge, we developed a computational technique that uses an image of a nuclear DNA stain and the rich morphology of the system to computationally label embryonic tissues at cellular resolution. The computational labeling learns to recognize a set of spatial identifiers that uniquely describe the morphology of a specific tissue in images of nuclear stained embryos annotated to reveal that tissue. The technique is then able to recognize the tissues in other unannotated embryos using only the nuclear image data. Once nuclei have been assigned to a specific tissue, we were then able to identify nuclear volumes using nuclear segmentation methods optimized for each tissue. This approach proved essential as the tissue optimized segmentation methods performed more accurately than a single generic method applied to all tissues equally. The resulting partial atlas allowed us to characterize the variation in cell numbers between embryos and left/right asymmetry in far more detail than previously possible and sets the stage for the development of a complete atlas. We will continue this work by 1. Establishing methods to recognize intricate tissues and cell shapes, 2. Building embryo atlases for three stages of development, and 3. Using live cell approaches to determine cell lineages between atlases of different stages.
Our project to create a quantitative, cellular resolution map of gene expression and morphology of embryogenesis will produce tools which will significantly advance basic biological research. Converting an animal embryo into a computational atlas of its body plan, cell positions, cell fate and tissue locations, along with a cellular resolution mapof gene expression, will provide a portable anatomical and expression atlas that can be dissected and observed by any researcher. This electronic embryonic atlas, will not only be an invaluable educational tool, but will allow cellular resolution anatomical analyses of anatomy in 3D, will shed light on the regulatory pathways, allow new types of computational systems biology and it will integrate anatomy with molecular developmental genetics.
|Barron, Jonathan T; ArbelÃ¡ez, Pablo; KerÃ¤nen, Soile V E et al. (2013) Volumetric Semantic Segmentation using Pyramid Context Features. Proc IEEE Int Conf Comput Vis 2013:3448-3455|
|Knowles, David W; Biggin, Mark D (2013) Building quantitative, three-dimensional atlases of gene expression and morphology at cellular resolution. Wiley Interdiscip Rev Dev Biol 2:767-79|
|Knowles, David W (2012) Three-dimensional morphology and gene expression mapping for the Drosophila blastoderm. Cold Spring Harb Protoc 2012:150-61|