Introduction

The completion of large-scale sequencing programmes has presented us with exciting challenges in how to determine the normal function of thousands upon thousands of genes. Fundamental to this understanding is where and when a gene is expressed. Such information provides clues into how a gene operates, and its interactions with other genes expressed in the same and neighbouring cells. For example, gene expression data is always collected in the analysis of mice bearing gene knockouts since discrete expression patterns are often predictive of the tissues likely to be affected by the absence of gene product. Gene expression patterns are also of high value to the biotechnology and pharmaceutical industries in the development of drug strategies (e.g. by defining the site of expression of pharmacological drug targets) and diagnostic tools (e.g. in designing predictive markers of clinical outcome), and may have increasing application in the identification and application of stem cell therapies.

Goal

Our goal over the time frame of this project (2003-2008) was to elucidate where and when large numbers of genes are expressed during development of the mouse embryo. While many complementing technologies are used to obtain gene expression data, we chose in situ hybridisation as a method to localise RNA transcripts in embryo sections. This approach has several advantages including the ability to explore gene expression at many levels from organs and tissues to individual cell types. In order to address questions of developmental mechanisms, it is imperative that expression patterns are determined in sufficient detail during development. We have adopted, therefore, a strategy of depth rather than breadth by producing detailed expression patterns of genes with likely to be of high developmental value over several developmental stages in the mouse embryo. This approach complements that of other larger high-throughput ‘atlas-type’ projects whose prim ary focus is many genes on one developmental stage or adult tissue. The high-resolution images of our data will be made freely available.

Methods

In outline, we use non-radioactive in situ hybridisation on embryo sections using an automated platform. Microscope slides are then scanned using a semi-automated imaging system, and the resulting images processed and deposited in our database for display over the internet. Unlike many high-throughput projects, we obtain very good tissue morphology, as we use 10 micron paraffin sections taken at regular intervals from paraformaldeyde-fixed embryos. We have developed highly sensitive techniques for in situ hybridisation that do not require a final signal amplification step. This may allow a more representative comparison between varying signal levels. These methods were made possible by adapting the Ventana Discovery automated platform to mouse embryo in situ hybridisation. Images from slides were created using the Applied Imaging Ariol scanning microscope. The Ariol captures individual frames that are registered together to create the full image (complete sections range in size from ~100 to 400MB). Individual frames are then taken from the microscope with Perl ImageMagick being used to stitch the individual tiles together. The VIPS image-processing library (University of Southampton) was employed to generate pyramidal tiff files that contain further sub-resolutions (i.e. zoom levels) of each image. The website uses the perl Catalyst web development framework. Pyramidal tiffs are served using
 IIPimage on a Fastcgi server working alongside 
Apache. On the client side, Javascript AJAX calls are made to the 
IIPserver to pull back tiles and assemble them in the browser. This allows us to efficiently serve the region of interest from extremely large files over the web.

Next...